Ironmaking power supply based on self-adaptive intelligent control

Abstract

The invention relates to the technical field of ironmaking power supply control, and discloses an ironmaking power supply based on self-adaptive intelligent control. A power supply working condition extraction module of the system carries out modeling according to a current load parameter and an operation state to obtain a power supply operation model; the historical operation strategy library forming module is used for layering historical regulation and control strategies according to historical performance parameter deviation and strategy confidence to form a layered strategy library; the core strategy generation module determines a minimum loss steady state strategy based on a target operation state, and screens a historical optimization strategy; the dynamic coordination module coordinates the two strategies to generate a control strategy through a double-channel mechanism; the actual state feedback module regulates output according to the control instruction and obtains an actual operation state; and the strategy library updating module updates the hierarchical strategy library according to the difference between the actual state and the target state. The power supply can improve the control adaptability and stability, adapt to the change of the ironmaking working condition and optimize the operation efficiency.

Description

Technical Field

[0001] This invention relates to the field of ironmaking power supply control technology, specifically to an ironmaking power supply based on adaptive intelligent control. Background Technology

[0002] In the modern steel industry production system, ironmaking, as a core process, places extremely high demands on the stability, precision, and energy efficiency of energy supply. Ironmaking power sources, as key equipment in energy supply, directly determine the product quality, production efficiency, and energy costs of the ironmaking process. With the steel industry's transformation towards "green, intelligent, and efficient" production, traditional ironmaking power sources are increasingly unable to meet the complex and ever-changing production demands, and a series of technological bottlenecks urgently need to be overcome.

[0003] Traditional ironmaking power supplies mostly employ fixed-parameter control or simple PID regulation, whose core drawback lies in their extremely poor adaptability to load fluctuations. During ironmaking, factors such as the reduction reaction of iron ore in the blast furnace, furnace temperature changes, and the uniformity of charge distribution cause frequent and drastic fluctuations in load parameters (such as current, voltage, and power factor), and fixed control strategies cannot respond to these dynamic changes in real time. When the load changes abruptly, the power supply output characteristics are prone to overshoot and oscillation, which not only causes process problems such as arc instability and uneven iron composition, but may also lead to power equipment failure due to instantaneous overload, seriously affecting the continuity and safety of ironmaking production.

[0004] From an energy consumption control perspective, traditional ironmaking power supplies lack refined optimization mechanisms for different operating conditions. During the ironmaking process, the power demand and output stability requirements vary significantly across different smelting stages (such as charging, melting, and refining). Existing control strategies often employ a single energy consumption control objective, failing to dynamically adjust control parameters based on specific operating conditions to achieve minimum energy loss. Statistics show that ironmaking power supplies using traditional control methods consume 15%-25% more energy than the theoretical optimal value, increasing energy costs for enterprises and severely contradicting current energy conservation and emission reduction requirements in the steel industry.

[0005] Traditional ironmaking power sources also have significant shortcomings in the iteration and optimization of control strategies. While some existing technologies incorporate historical operating data reference mechanisms, they lack a scientific strategy hierarchy and confidence assessment system, resulting in extremely low reusability and effectiveness of historical strategies. On one hand, historical control strategies are often stored haphazardly, making it difficult to quickly select optimized strategies that match the current state when new operating conditions need to be addressed, leading to low decision-making efficiency. On the other hand, the lack of quantitative evaluation of the effectiveness of historical strategies fails to consider the degree of deviation and applicability of different strategies under different operating conditions. Direct reuse may lead to counterproductive control effects or even introduce new operational risks.

[0006] Traditional feedback control mechanisms for ironmaking power supplies suffer from lag and incompleteness. Existing feedback systems mostly collect basic parameters such as power output voltage and current, lacking in-depth analysis of the correlation between actual smelting conditions (such as furnace temperature distribution and material reaction progress) and the power supply's operating status. This results in an inability to accurately identify key discrepancies affecting control effectiveness. When deviations occur between the actual operating state and the target state, adjustment commands are often generated based on a single parameter feedback, failing to fundamentally resolve the deviation problem and creating a vicious cycle of "adjustment-deviation-readjustment," further reducing the accuracy and stability of power supply control.

[0007] With the deep application of technologies such as the Industrial Internet and artificial intelligence in the steel industry, ironmaking production places higher demands on the intelligence level of power supply control. The industry urgently needs a new ironmaking power supply control technology capable of adaptive perception of operating conditions, dynamic strategy optimization, and precise energy consumption control. This technology aims to overcome the limitations of traditional technologies, improve the operational stability, energy efficiency, and intelligence of ironmaking power supplies, and provide key technological support for the green and low-carbon transformation of the steel industry. Against this backdrop, developing an ironmaking power supply based on adaptive intelligent control has significant practical implications and industrial value. Summary of the Invention

[0008] The purpose of this invention is to provide an ironmaking power source based on adaptive intelligent control to solve the problems mentioned in the background art.

[0009] To achieve the above objectives, the present invention provides an ironmaking power supply based on adaptive intelligent control, the system comprising:

[0010] The power supply condition extraction module is used to model the current load parameters and operating status of the ironmaking power supply in order to obtain the power supply operation model.

[0011] The historical operation strategy library formation module is used to divide the historical control strategies of the ironmaking power source into multiple levels based on the deviation of historical performance parameters and the strategy confidence level, so as to form a hierarchical strategy library.

[0012] The core strategy generation module is used to determine the lowest loss steady-state strategy to achieve the target operating state based on the target operating state of the power supply in the set power supply operation model, and to filter the hierarchical strategy library based on the strategy confidence and state matching degree to obtain historical optimization strategies.

[0013] The dynamic coordination module is used to dynamically coordinate the steady-state strategy and the historical optimization strategy using a dual-path mechanism, thereby generating a power control strategy.

[0014] The actual status feedback module is used to adjust the power output characteristics based on power control commands in order to obtain the actual operating status of the power supply, wherein the power control commands are generated based on the power control strategy and the current load parameters.

[0015] The strategy library update module is used to determine the factors that cause the difference between the actual running state and the target running state, and to update the content of the hierarchical strategy library based on the factors that cause the difference.

[0016] Preferably, the historical operation strategy library formation module includes:

[0017] The strategy evaluation unit identifies the deviation between the confidence level of each control strategy and its historical performance parameters based on historical operating data.

[0018] The hierarchical division unit divides the historical control strategy into a high-confidence strategy layer, a medium-confidence strategy layer, and a basic strategy layer based on the confidence level and the deviation, to form a hierarchical strategy library.

[0019] Preferably, the core strategy generation module includes:

[0020] The load spectrum analysis unit uses a sensor system to collect power load fluctuation data and power quality data.

[0021] The steady-state strategy generation unit inputs the load fluctuation data and power quality data into the intelligent analysis model, performs calculations using the target operating state as the optimization objective, and outputs the lowest-loss steady-state strategy to achieve the target operating state.

[0022] Preferably, the core strategy generation module further includes:

[0023] The strategy sample extraction unit extracts reference strategy samples from each level of the hierarchical strategy library based on the strategy confidence level.

[0024] The historical strategy matching unit filters the reference strategy samples based on the state matching degree to obtain historical optimization strategies.

[0025] Preferably, the dynamic coordination module includes:

[0026] The hysteresis feature analysis unit extracts the hysteresis features of the power supply response based on historical load response data and generates the load response hysteresis.

[0027] The nonlinear interference analysis unit collects power supply output parameters when the load changes abruptly, determines the nonlinear interference quantity based on the coupling characteristics of voltage and current, and determines the power supply's anti-interference robustness by combining the correlation between voltage and current fluctuations.

[0028] Preferably, the dynamic coordination module further includes:

[0029] The dual-path coordination mechanism monitors the load change rate and nonlinear interference in real time. When the change rate exceeds the threshold, the historical optimization strategy path is activated; otherwise, the steady-state strategy path is maintained, and the power control strategy is generated by fusion.

[0030] Preferably, the actual state feedback module includes:

[0031] The multi-scale feature extraction unit performs multi-scale decomposition on the power supply output waveform to obtain current characteristics in different frequency bands;

[0032] The regulation feature decoupling unit processes the current features independently, generating feature vectors for waveform quality classification and feature vectors for output regulation.

[0033] Preferably, the actual state feedback module further includes:

[0034] The operating status quantization unit analyzes the actual output waveform distortion rate and energy efficiency parameters of the power supply based on the decoupled feature vectors to generate the actual operating status of the power supply.

[0035] Preferably, the strategy library update module includes:

[0036] The state difference diagnosis unit compares the waveform distortion difference value and energy efficiency difference value between the actual operating state and the target operating state;

[0037] The key factor identification unit locates the load fluctuation stage based on the difference value and identifies the power regulation factors that cause the difference;

[0038] The strategy library reconstruction unit updates the strategy parameters of the corresponding level in the hierarchical strategy library according to the power control factors.

[0039] Preferably, the strategy library update module further includes:

[0040] The confidence recalibration unit dynamically adjusts the policy confidence of each level in the hierarchical policy library based on the performance deviation of the updated policy in subsequent operations.

[0041] Compared with the prior art, the beneficial effects of the present invention are:

[0042] This ironmaking power supply based on adaptive intelligent control, through the setting of a power supply condition extraction module, can construct a power supply operation model based on the current load parameters and operating status. This makes the power supply control more closely aligned with real-time operating conditions, avoiding the problem of insufficient adaptability caused by relying on preset parameters in traditional fixed modes. Through precise modeling, the current operating characteristics of the power supply can be clearly understood, providing a basis for generating subsequent control strategies that meet actual needs and making the control process more targeted.

[0043] The historical operation strategy library module categorizes historical control strategies into multiple levels based on historical performance parameter deviations and strategy confidence levels, constructing a hierarchical strategy library that fully leverages the value of historical operation data. This hierarchical approach effectively distinguishes strategies with different confidence levels, allowing for rapid focus on highly adaptable historical optimization strategies during subsequent strategy selection. This reduces the time spent filtering ineffective strategies and provides diverse references for strategy selection under different operating conditions, avoiding the limitations of a single strategy in complex situations.

[0044] The core strategy generation module simultaneously determines the minimum loss steady-state strategy and the historical optimization strategy, achieving a combination of theoretically optimal and practically optimized strategies. The minimum loss steady-state strategy, from the perspective of energy consumption control, ensures the economy of the power supply under the target operating state; the historical optimization strategy, based on past effective control experience, ensures the feasibility and effectiveness of the strategy in practical applications. The two work together to provide more comprehensive strategy support for power supply control, reducing the potential one-sidedness of a single strategy.

[0045] The dynamic coordination module employs a dual-path mechanism to coordinate steady-state and historical optimization strategies, flexibly adjusting the weight distribution of the two strategies based on the real-time operating status of the power supply. When the power supply is in a stable operating condition, the steady-state strategy can be emphasized to maintain low-loss operation; when the operating condition fluctuates or approaches a complex state, the weight of the historical optimization strategy can be increased to improve the strategy's responsiveness to changes in operating conditions, avoid the response lag problem of a single strategy during operating condition switching, and enhance the stability of power supply operation.

[0046] The actual status feedback module adjusts the output characteristics based on power control commands and acquires the actual operating status, forming a closed-loop control-feedback process. By acquiring the actual operating status in real time, the module can promptly perceive the implementation effect of the control strategy, understand the difference between the actual operating status and the target status of the power supply, and provide a direct basis for subsequent strategy optimization and library updates. This prevents the control process from deviating from the actual operating situation and ensures that the adjustment direction always aligns with the requirements.

[0047] The strategy library update module updates the hierarchical strategy library based on the differences between the actual and target operating states, enabling the strategy library content to be dynamically optimized in line with actual operating results. This real-time update mechanism breaks through the limitations of traditional manual periodic maintenance, ensuring that the strategy library always maintains a high degree of adaptability. As operating time progresses, the proportion of effective strategies in the strategy library continuously increases, and the accuracy of subsequent strategy selection also improves. This, in turn, drives the continuous optimization of the adaptability and accuracy of power supply control, helping ironmaking power supplies maintain a high-efficiency and stable operating state in the long term, adapting to the changing operating conditions in the ironmaking process. Attached Figure Description

[0048] Figure 1 This is a schematic diagram illustrating the working principle of the ironmaking power supply based on adaptive intelligent control as described in this invention.

[0049] Figure 2 A flowchart illustrating the working principle of the module forming the historical operation strategy library;

[0050] Figure 3 A flowchart illustrating the working principle of the core strategy generation module;

[0051] Figure 4 This is a flowchart illustrating the working principle of the dynamic coordination module. Detailed Implementation

[0052] 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.

[0053] Please see Figure 1 This invention provides an ironmaking power supply based on adaptive intelligent control, the system comprising:

[0054] The power supply condition extraction module models the current load parameters and operating status of the ironmaking power supply. It constructs a power supply operation model by collecting real-time data such as voltage, current, and power factor. This model uses state-space representation to describe the power supply's dynamic characteristics. The historical operation strategy library formation module processes historical control strategies, classifying them hierarchically based on the deviation of historical performance parameters and strategy confidence, forming a hierarchical strategy library. Strategy confidence is calculated based on the success rate and stability of historical strategies. The core strategy generation module, based on the target operating state set in the power supply operation model, determines the lowest-loss steady-state strategy to achieve the target state through optimization algorithms. Simultaneously, it selects historical optimized strategies from the hierarchical strategy library based on strategy confidence and state matching. The dynamic coordination module uses a dual-path mechanism to dynamically coordinate the steady-state strategy and historical optimized strategies, generating the final power supply control strategy. This mechanism switches strategy paths based on real-time load changes and nonlinear disturbances. The actual state feedback module adjusts the power supply output characteristics based on power control commands, obtaining actual operating state data of the power supply by adjusting power output and waveform parameters. The strategy library update module compares the actual operating state with the target operating state, identifies the difference factors such as waveform distortion or energy efficiency deviation, and updates the strategy parameters and confidence levels in the hierarchical strategy library based on these factors, thereby completing the adaptive control loop.

[0055] Example 1: See Figure 2In the adaptive intelligent control system of ironmaking power supply, the formation of a historical operation strategy database is a core component. This database uses a systematic approach to extract, evaluate, and classify past control experiences, thereby constructing a hierarchical and highly efficient knowledge base. Assume that the power supply system of an ironmaking plant has been operating continuously for many years, and its data center stores massive amounts of historical operation data records. Each record contains, at a specific point in time, the load parameters faced by the power supply (such as load current magnitude, load type (resistive, inductive, or capacitive), load transient rate, etc.), the control strategies adopted (such as adjusting the output voltage amplitude, changing the phase, switching the PWM modulation mode, etc.), and the performance parameters obtained after executing the strategy (such as final output stability, power conversion efficiency, total harmonic distortion, etc.).

[0056] The strategy evaluation unit first preprocesses this historical data, removing invalid and outlier data points, and then begins a quantitative evaluation of each historically employed control strategy. The evaluation focuses on two core indicators: the deviation of historical performance parameters and strategy confidence. Taking a control strategy record for "high current surge during the initial startup of an electric arc furnace" as an example, the deviation of the historical performance parameters of this strategy is calculated by comparing the actual output after adopting the strategy with the expected ideal output. For example, the expected goal is to control voltage fluctuations within ±5%, but in actual operation, the strategy may have repeatedly controlled fluctuations within ±8%. This deviation value is quantified using statistical methods (such as calculating the root mean square error). Meanwhile, the calculation of strategy confidence is more comprehensive, considering the ratio of the number of times the strategy was successfully executed to the total number of executions (i.e., the success rate), and the stability of the strategy under different but similar operating conditions. For example, if the above strategy keeps the fluctuations within an acceptable range in eight out of ten similar electric arc furnace start-up events, its success rate is high; if the performance results (fluctuation range) of these eight successful executions are very similar, its stability is high. The combination of the two gives the strategy a high confidence score.

[0057] After evaluating all historical strategies, the hierarchical partitioning unit begins its work. Based on the calculated confidence scores and deviation values, it maps all strategies onto a two-dimensional coordinate system, with the horizontal axis representing deviation and the vertical axis representing confidence. Subsequently, a clustering algorithm (such as K-means clustering) is used to naturally aggregate these data points into three groups. The first group contains strategies with extremely high confidence scores and extremely low deviation values. These strategies are often well-validated optimization schemes used under mature operating conditions, demonstrating reliable and excellent performance. They are categorized as the "high-confidence strategy layer." For example, the constant power control strategy for the "steady-state heating stage of the blast furnace," due to its stable load, singular control objective, and historical data showing high success rate and excellent output waveform quality, is selected for this layer.

[0058] The second group contains strategies with moderate confidence and bias. These strategies may perform well in some aspects but are inadequate in others, or their effectiveness is heavily dependent on boundary conditions. They are classified as the "medium confidence strategy layer." For example, a fuzzy adaptive PID control strategy for "slight load fluctuations" can suppress fluctuations, but its settling time can sometimes be too long, and its effect is not as immediate and stable as that of high confidence strategies; therefore, it is classified into this layer.

[0059] The third group includes strategies with low confidence levels, large biases, or insufficient application data. These strategies may be experimental solutions for extreme or rare operating conditions, and their effectiveness needs further validation. They are categorized as the "basic strategy layer" or the "candidate strategy layer." For example, one record shows a composite control strategy tried to address a complex fault where "a sudden drop in grid voltage occurs simultaneously with a sudden increase in load." Because the probability of this composite event is extremely low, the strategy was only executed once or twice, and the data is insufficient to prove its general effectiveness; therefore, it was placed in the basic layer.

[0060] The resulting hierarchical strategy library is physically represented as a structured database. Strategies in the high-confidence strategy layer have the highest invocation priority and are typically the first reference for the core strategy generation module. The medium-confidence strategy layer serves as an effective supplement and alternative. The basic strategy layer provides a valuable knowledge base to handle rare operating conditions and offers training samples for machine learning algorithms. The continuous operation of this process allows the strategy library to constantly absorb new operational experience, be optimized and adjusted through subsequent update modules, becoming a dynamically evolving "brain" that supports the continuous enhancement of the adaptability of the intelligent control system for ironmaking power supplies.

[0061] Example 2: See Figure 3In the operation of the adaptive intelligent control system for ironmaking power supply, the core strategy generation module plays a crucial role in bridging the gap between upstream and downstream processes. On the one hand, it needs to generate optimal steady-state strategies for the future based on real-time load perception; on the other hand, it needs to draw wisdom from historical experience to select optimization strategies that can be referenced. Assume that the electric arc furnace in the main ironmaking process is in the middle of the smelting process, a typical stage of intense and frequent load fluctuations. At this time, the load spectrum analysis unit begins continuous operation. The sensor system installed in the power output circuit and grid connection point, including high-precision current transformers, voltage sampling circuits, and dedicated power quality analyzers, begins to capture data at an extremely high sampling rate. They record the instantaneous changes in load current, whose waveforms exhibit the unique random and abrupt changes characteristic of the electric arc furnace; simultaneously, voltage fluctuations, minor frequency shifts, and a large number of harmonic components generated by the nonlinear characteristics of the electric arc are also recorded synchronously. These raw "load fluctuation data" and "power quality data," after preliminary filtering and formatting, are converted into a structured data stream containing multi-dimensional features such as timestamps, amplitudes, phases, and harmonic spectra, and transmitted to subsequent units in real time.

[0062] The steady-state strategy generation unit is the core of the forward-looking computation. It receives data streams from the load spectrum analysis unit and inputs them into a pre-trained intelligent analysis model. Internally, this model employs a deep neural network architecture, trained on massive amounts of historical and simulation data. Its learning objective is to establish a complex nonlinear mapping between "load and power quality inputs" and "optimal power supply control parameter outputs." Currently, the unit's "target operating state" is to maintain the output voltage stable at its rated value within a future control cycle, while minimizing the switching and conduction losses of power devices such as IGBTs within the power supply. The intelligent model performs deep computation on the input multi-dimensional feature data, simulating the output results under different control strategies, and ultimately outputs a specific set of control parameters, such as the optimal switching frequency, duty cycle adjustment, and phase compensation value. This set of parameters constitutes the "lowest-loss steady-state strategy" for achieving the target operating state, representing the theoretically optimal solution under the current load forecast.

[0063] The strategy sample extraction unit starts in parallel, accessing the "hierarchical strategy library" built by the historical strategy library formation module. This unit does not blindly call all strategies, but rather extracts strategy samples from the high-confidence, medium-confidence, and basic strategy layers according to a certain proportion and weight, based on the quantitative indicator of "strategy confidence" attached to each historical strategy in the database. For example, it might extract the 5 highest-confidence strategies from the high-confidence layer, 10 representative strategies from the medium-confidence layer, and 3 as exploratory samples from the basic layer. This confidence-based sampling method ensures that the extracted sample set includes both well-validated reliable strategies and alternative solutions that may be applicable to special cases, forming a diverse pool of "reference strategy samples."

[0064] The historical strategy matching unit is responsible for fine-tuning these reference samples. Its core task is to calculate the "state matching degree" between each reference strategy sample and the current real-time operating conditions. The matching degree calculation is not a simple data comparison, but rather a comprehensive consideration of the similarity between the load characteristics and power quality environment when the historical strategy was applied and the current situation. For example, it analyzes whether the dominant frequency of the current load fluctuation matches the characteristics in a historical strategy record, and whether the current harmonic distortion mode is similar to another record. Through a series of pattern recognition and similarity calculation algorithms, a matching degree score is calculated for each reference strategy sample. Finally, the historical strategies with the highest matching degree scores are selected to form a "historical optimized strategy" set. These strategies represent solutions that have proven effective under similar historical operating conditions.

[0065] The core strategy generation module produced two policy outcomes: one is a theoretically calculated "lowest-loss steady-state strategy" optimized for the future; the other is a "historical optimized strategy" derived from past experience and selected through matching. These two sets of strategies were fed together into the dynamic coordination module, preparing for the final generation of a power control strategy that can adapt to future changes while drawing on historical best practices. The coordinated operation of this entire process demonstrates the control system's profound understanding and adaptive capabilities in complex industrial environments.

[0066] Example 3: See Figure 4In an ironmaking power supply system based on adaptive intelligent control, the dynamic coordination module is responsible for fusing and deciding on the steady-state strategy and historical optimization strategy provided by the core strategy generation module. Its operation mechanism is based on a deep understanding and real-time monitoring of the system's dynamic characteristics. The operation of the dynamic coordination module begins with the hysteresis characteristic analysis unit. This unit does not process real-time data but performs offline analysis on accumulated historical load response data, aiming to extract the inherent dynamic response characteristics of the power supply system. This historical data records the complete trajectory of the power supply's output parameters (such as voltage and current) changing over time during countless load change events. By systematically analyzing these trajectories, such as by applying system identification methods, this unit can quantify the inherent delay in the power supply response. This delay may stem from various factors, including the switching characteristics of power devices, the data processing time of the control loop, and the phase lag of the output filter. The analysis results are summarized into a comprehensive index called load response hysteresis (HH). This time-dimensional parameter describes the typical time interval between identifying a load change and the power supply output completing its adjustment. The hysteresis is calculated based on a statistical fit of historical data and can be expressed as:

[0067]

[0068] in: This represents the total number of historical load mutation events analyzed. Indicates the first The moment when the load of this event begins to change abruptly. Then it means the first The moment when the power supply output returns to a stable range after this event. This hysteresis. This is an important prior knowledge used to predict the system's response speed to the current disturbance.

[0069] The nonlinear interference analysis unit focuses on processing real-time data, especially activating immediately upon detecting a sudden change in load. When the load current changes abruptly, this unit rapidly acquires the output voltage and current waveforms of the power supply. Due to the strong nonlinear characteristics of ironmaking loads (such as electric arc furnaces), voltage and current changes are not independent but rather exhibit complex coupling relationships. This coupling manifests as follows: a sudden change in current causes a sharp drop or rise in voltage, and the voltage change, in turn, affects the equivalent impedance of the arc, further altering the current. This unit estimates a quantity called nonlinear interference by analyzing the interaction between the voltage and current waveforms at the moment of the sudden change. The dimensionless parameter of the current disturbance (VFD) characterizes the strength of the system's nonlinearity under the current disturbance. Furthermore, by analyzing the strength of the correlation between voltage and current fluctuations, the system's ability to suppress this mutual interference can be assessed, i.e., its robustness against disturbances. In a robust system, voltage and current fluctuations are relatively independent; conversely, weakly robust systems are more susceptible to nonlinear coupling.

[0070] The dual-path coordination mechanism is the core of the module's decision-making; it receives the outputs (load response hysteresis) from the two units mentioned above in real time. and real-time nonlinear disturbance quantity ), while continuously monitoring the current load change rate ( The mechanism internally sets a rate of change threshold (). This threshold is provided based on the rated capacity of the power supply unit, its historical operating performance, and the hysteresis characteristic analysis unit. This is a comprehensive setting. Its decision-making logic is as follows: The system continuously calculates the absolute value of the current load change rate. .like If the system is currently in a normal fluctuation range, it will maintain the "steady-state strategy path". That is, it will directly use the lowest loss steady-state strategy calculated by the core strategy generation module as the current control command, because this strategy has the best economy under stable operating conditions.

[0071] If detected This indicates a drastic change in load, exceeding the optimization range of the steady-state strategy. At this point, the mechanism immediately activates the "historical optimization strategy path." It will incorporate the current load change characteristics and nonlinear disturbances... and lag As query criteria, one or more strategies with the highest matching degree are quickly selected from the historical optimized strategy set provided by the core strategy generation module. The final control strategy is not a simple binary choice, but a fusion generation. The fusion algorithm assigns high weight to historical optimized strategies because they are derived from experience verified under similar severe disturbances. At the same time, it also refers to the long-term optimization objectives of steady-state strategies, making fine adjustments and corrections to generate a power control strategy that can quickly respond to the current instantaneous disturbance while taking into account long-term operational economy. This dual-path design enables the system to maintain high efficiency and stability under normal conditions, while also possessing agility and reliability in dealing with sudden disturbances.

[0072] Example 4: In an ironmaking power supply system based on adaptive intelligent control, the actual state feedback module plays a crucial role in translating control strategies into actual actions and accurately evaluating their effects. After the power control strategy is generated and issued by the dynamic coordination module, the power conversion unit executes the corresponding control commands, adjusting the switching modes of its power devices to change the final power output characteristics. At this time, the actual state feedback module begins operation, its primary task being to deeply perceive and analyze the actual output waveform of the power supply. The multi-scale feature extraction unit, acting as a front-end processor, synchronously captures the voltage and current waveforms at the power output terminal through a high-precision data acquisition card. Taking the complex changes in the power output current when the electric arc furnace load enters the refining stage as an example, this unit uses multi-scale signal analysis technology to process the original waveform. It does not treat the waveform as a whole but decomposes it into different time-frequency scales. Through a series of filtering and decomposition algorithms, the original composite waveform is separated into different components that constitute its details. For example, a complex current signal containing high-frequency glitches, mid-frequency fluctuations, and low-frequency trends is decomposed into multiple sub-signals of different frequency bands. Each sub-signal carries information with different physical meanings: the high-frequency band may reflect the switching noise of the power devices and the resulting higher harmonics; the mid-frequency band may contain the main operating frequency of the load and the main harmonic components in its vicinity; and the low-frequency band may reflect the overall trend of load power variation. The current characteristics of these separated frequency bands provide a clear data foundation for subsequent independent analysis.

[0073] The decoupling unit for control features receives these multi-scale current characteristic data. The core design philosophy of this unit recognizes that different components of the waveform have different uses and meanings and need to be treated differently. Therefore, this unit establishes parallel, independent processing channels to decouple these features. One processing channel is dedicated to evaluating power quality. It focuses on extracting parameters related to waveform purity from mid-to-high frequency characteristics, such as the amplitude of each harmonic, a preliminary estimate of the total harmonic distortion rate, and waveform symmetry. These parameters are calculated and integrated to construct a feature vector for waveform quality classification. This vector acts like a "quality report," quantitatively describing the deviation of the current output waveform from the ideal sine wave. The other processing channel focuses on analyzing power transmission status. It mainly extracts information from low-frequency trend components and the fundamental component, calculating parameters such as active power, reactive power, apparent power, power factor, and dynamic response speed. These parameters are integrated into a feature vector for output control, which acts more like a "performance report," describing the power supply's energy delivery efficiency and load following ability. This decoupling process makes the evaluation dimensions of the power supply state clearer and avoids mutual interference between different characteristics.

[0074] The operational state quantization unit is the final step in generating a usable state description. It receives the two decoupled feature vectors mentioned above, performs comprehensive calculations and fusions on them, and ultimately generates a quantitative description of the actual operating state of the power supply. This state description is a structured dataset, containing two core types of parameters: one is waveform quality parameters, the core indicator of which is the precisely calculated waveform distortion rate, which integrates the effects of all harmonics and interharmonics; the other is energy efficiency parameters, including overall conversion efficiency, power factor, etc. These parameters together constitute an objective and quantitative evaluation of the "performance of the power supply control strategy". To more intuitively demonstrate the output results of the multi-scale feature extraction unit, refer to Table 1, which lists exemplary data of the frequency band features after current waveform decomposition within one analysis cycle.

[0075] Table 1: Multiscale Current Characteristic Decomposition Table.

[0076] Frequency band allocation Center frequency (Hz) Feature description The main physical phenomena reflected low frequency band 50 The fundamental current amplitude is stable, but exhibits slow drift. Overall demand for active power of load and slow changes Mid-low frequency band 150 The third harmonic component is significant, and its amplitude fluctuates. Characteristic harmonics generated by nonlinear loads of electric arc furnaces Mid-to-high frequency band 1050 There are multiple harmonic groups with low amplitude. Harmonic radiation near the switching frequency of power devices High frequency band 10000 Wideband noise with very small amplitude High-frequency electromagnetic noise caused by switching action

[0077] Example 5: In an ironmaking power supply system based on adaptive intelligent control, the strategy library update module is the core component for achieving continuous system evolution and self-improvement. This module performs in-depth comparative analysis of the actual operating state data of the power supply provided by the actual state feedback module with the preset target operating state. The operation of the strategy library update module begins with the state difference diagnosis unit. This unit receives quantitative data from the actual state feedback module, which accurately describes the actual output performance of the power supply under the latest control strategy, including waveform distortion rate and multiple energy efficiency parameters. At the same time, this unit retrieves the target operating state parameters set in the current operating cycle, which is the expected result of the previous calculation by the core strategy generation module. The diagnosis unit performs a detailed comparison of these two sets of data item by item. The calculation process is not a simple subtraction, but uses a weighted algorithm to comprehensively consider the physical meaning of each parameter and its impact on the overall system performance, and finally outputs a set of quantitative difference values, the most important of which are the waveform distortion difference value and the energy efficiency difference value. These difference values ​​not only reflect the difference in absolute values, but sometimes also contain trend information, such as the distortion rate continuously increasing within a certain period of time.

[0078] The key factor identification unit then conducts in-depth source analysis based on these discrepancies. Its primary task is to accurately pinpoint the specific time interval causing the discrepancies, i.e., to identify the "load fluctuation phase." It performs correlation analysis between high-precision time-stamped data and the historical records of the load spectrum analysis unit, seeking the temporal correspondence between peak discrepancy values ​​and specific load change patterns (such as sudden rises and falls of electric arc furnace electrodes, or collapse of scrap steel). After successfully locating the problem period, the identification unit further analyzes all operational data within that period, including power supply control command sequences, power device response logs, and instantaneous load characteristics. Through a series of pattern recognition and causal reasoning algorithms, this unit strives to sift through massive amounts of data to identify the key causes of performance deviations, i.e., "power supply control factors." These factors may be very specific; for example, when the load current rises sharply, the voltage loop proportional gain parameter in the adopted strategy may be set too low, resulting in insufficient dynamic response and thus causing a large voltage sag and subsequent waveform distortion; or, when dealing with a specific number of harmonics, the filter parameters set in the strategy may fail to achieve optimal reduction.

[0079] The strategy base reconstruction unit executes specific knowledge base update operations based on the conclusions of the key factor identification unit. This unit accesses the hierarchical strategy base and first determines the level (high confidence, medium confidence, or basic strategy layer) of the strategy that needs updating based on the type and source of the identified regulatory factors. For example, if a strategy originally belonging to the high confidence strategy layer is found to have a defect in a certain parameter under specific operating conditions, the reconstruction unit will locate that strategy. The update is not simply replacing the old value with a new one, but a careful correction process. It may add new applicable constraints to the strategy, specify the effective range of its parameters; or it may directly fine-tune and optimize specific parameters in the strategy to make it more suitable for the identified operating conditions. For strategies in the basic layer, the update may manifest as supplementing new annotation information or related data to enrich its knowledge dimensions.

[0080] The confidence recalibration unit is responsible for maintaining the dynamic accuracy of metadata in the policy repository. This unit continuously monitors the performance of all policies, especially those that have been updated or invoked, in subsequent actual operations. It tracks the performance deviations of these policies after each execution, recording their successes and failures. Based on this new historical data, the unit uses a recursive algorithm to periodically recalculate the confidence score of each policy. A policy whose performance improves after being updated may have its confidence score increased; conversely, a policy that repeatedly fails to achieve the expected results, even if it was originally in a high-confidence layer, will have its confidence score automatically decreased by the system, potentially changing its hierarchical position in the hierarchical policy repository. This dynamic recalibration mechanism ensures that the policy repository's rating system always reflects the current true reliability of the policies, making the entire hierarchical policy repository a vibrant knowledge ecosystem capable of continuously self-optimizing as the operating environment changes.

[0081] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0082] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A smelting power supply based on adaptive intelligent control, characterized in that, include: The power supply condition extraction module is used to model the current load parameters and operating status of the ironmaking power supply in order to obtain the power supply operation model. The historical operation strategy library formation module is used to divide the historical control strategies of the ironmaking power source into multiple levels based on the deviation of historical performance parameters and the strategy confidence level, so as to form a hierarchical strategy library. The core strategy generation module is used to determine the lowest loss steady-state strategy to achieve the target operating state based on the target operating state of the power supply in the set power supply operation model, and to filter the hierarchical strategy library based on the strategy confidence and state matching degree to obtain historical optimization strategies. The dynamic coordination module is used to dynamically coordinate the steady-state strategy and the historical optimization strategy using a dual-path mechanism, thereby generating a power control strategy. The actual status feedback module is used to adjust the power output characteristics based on power control commands in order to obtain the actual operating status of the power supply, wherein the power control commands are generated based on the power control strategy and the current load parameters. The strategy library update module is used to determine the factors that cause the difference between the actual running state and the target running state, and to update the content of the hierarchical strategy library based on the factors that cause the difference.

2. The ironmaking power supply based on adaptive intelligent control according to claim 1, characterized in that, The historical operation strategy library formation module includes: The strategy evaluation unit identifies the deviation between the confidence level of each control strategy and its historical performance parameters based on historical operating data. The hierarchical division unit divides the historical control strategy into a high-confidence strategy layer, a medium-confidence strategy layer, and a basic strategy layer based on the confidence level and the deviation, to form a hierarchical strategy library.

3. The ironmaking power supply based on adaptive intelligent control according to claim 2, characterized in that, The core strategy generation module includes: The load spectrum analysis unit uses a sensor system to collect power load fluctuation data and power quality data. The steady-state strategy generation unit inputs the load fluctuation data and power quality data into the intelligent analysis model, performs calculations using the target operating state as the optimization objective, and outputs the lowest-loss steady-state strategy to achieve the target operating state.

4. The ironmaking power supply based on adaptive intelligent control according to claim 3, characterized in that, The core strategy generation module also includes: The strategy sample extraction unit extracts reference strategy samples from each level of the hierarchical strategy library based on the strategy confidence level. The historical strategy matching unit filters the reference strategy samples based on the state matching degree to obtain historical optimization strategies.

5. The ironmaking power supply based on adaptive intelligent control according to claim 4, characterized in that, The dynamic coordination module includes: The hysteresis feature analysis unit extracts the hysteresis features of the power supply response based on historical load response data and generates the load response hysteresis. The nonlinear interference analysis unit collects power supply output parameters when the load changes abruptly, determines the nonlinear interference quantity based on the coupling characteristics of voltage and current, and determines the power supply's anti-interference robustness by combining the correlation between voltage and current fluctuations.

6. The ironmaking power supply based on adaptive intelligent control according to claim 5, characterized in that, The dynamic coordination module also includes: The dual-path coordination mechanism monitors the load change rate and nonlinear interference in real time. When the change rate exceeds the threshold, the historical optimization strategy path is activated; otherwise, the steady-state strategy path is maintained, and the power control strategy is generated by fusion.

7. The ironmaking power supply based on adaptive intelligent control according to claim 6, characterized in that, The actual status feedback module includes: The multi-scale feature extraction unit performs multi-scale decomposition on the power supply output waveform to obtain current characteristics in different frequency bands; The regulation feature decoupling unit processes the current features independently, generating feature vectors for waveform quality classification and feature vectors for output regulation.

8. The ironmaking power supply based on adaptive intelligent control according to claim 7, characterized in that, The actual status feedback module also includes: The operating status quantization unit analyzes the actual output waveform distortion rate and energy efficiency parameters of the power supply based on the decoupled feature vectors to generate the actual operating status of the power supply.

9. The ironmaking power supply based on adaptive intelligent control according to claim 8, characterized in that, The policy library update module includes: The state difference diagnosis unit compares the waveform distortion difference value and energy efficiency difference value between the actual operating state and the target operating state; The key factor identification unit locates the load fluctuation stage based on the difference value and identifies the power regulation factors that cause the difference; The strategy library reconstruction unit updates the strategy parameters of the corresponding level in the hierarchical strategy library according to the power control factors.

10. The ironmaking power supply based on adaptive intelligent control according to claim 9, characterized in that, The policy library update module also includes: The confidence recalibration unit dynamically adjusts the policy confidence of each level in the hierarchical policy library based on the performance deviation of the updated policy in subsequent operations.

Images