Hybrid energy storage thermal power unit frequency modulation method based on hierarchical adaptive control

Abstract

The application relates to the technical field of power system frequency modulation control, and specifically discloses a hybrid energy storage thermal power unit frequency modulation method based on hierarchical adaptive control, which extracts a feature sequence from real-time collected throttle stroke data, battery operation data, super capacitor charge data and frequency modulation instructions, on the basis of which, mechanical fatigue loss evolution of the cumulative stroke of the throttle is carried out, and the discharge depth of the battery is quantified in terms of microcirculation equivalent capacity loss to obtain fatigue degree indexes and attenuation degree indexes, which are then mapped to a two-dimensional health situation coordinate and subjected to potential field optimization deduction by a fuzzy logic reasoning engine to generate an adjustment coefficient vector; the vector is used in combination with the energy storage state of charge to adaptively update the filter cutoff frequency and the power distribution dead zone; and finally, the frequency modulation instruction is subjected to dead zone truncation filtering and hierarchical frequency domain decoupling based on the updated parameters, and the power instructions of the thermal power, the battery and the super capacitor are output.

Description

Technical Field

[0001] This application relates to the field of frequency regulation control technology for power systems, and more specifically, to a frequency regulation method for hybrid energy storage thermal power units based on hierarchical adaptive control. Background Technology

[0002] With the large-scale grid connection of high-proportion renewable energy sources, the rotational inertia of the power system continues to decrease, and the amplitude and frequency of grid frequency fluctuations have increased significantly, placing higher demands on the response speed and regulation accuracy of automatic generation control frequency regulation services. Traditional thermal power units are constrained by the inherent large inertia and long time constant of the boiler-turbine thermal system, resulting in shortcomings such as response lag and insufficient ramp rate when facing high-frequency random frequency regulation commands, making it difficult to independently meet frequency regulation assessment standards. While relying solely on electrochemical energy storage or supercapacitors can achieve rapid power response, their limited energy capacity and high life-cycle costs also prevent them from independently undertaking large-scale frequency regulation tasks for extended periods. Therefore, power complementary coupling between thermal power units and hybrid energy storage systems composed of lithium batteries and supercapacitors to construct a hierarchical collaborative frequency regulation architecture has become the mainstream technical approach for improving frequency regulation performance and reducing overall costs.

[0003] Existing hybrid energy storage and thermal power joint frequency regulation schemes generally adopt hierarchical decoupling control based on signal frequency boundaries. This involves using high-pass and low-pass filters with fixed parameters or adaptive adjustment based on the energy storage state of charge to break down frequency regulation commands into power sub-commands for different frequency bands, which are then distributed to the thermal power unit, lithium battery, and supercapacitor for execution. However, the adaptive adjustment of these schemes is limited to the single electrical dimension of the energy storage state of charge, failing to incorporate the mechanical fatigue wear of the turbine control valve and the electrochemical micro-cycle lifespan decay of the lithium battery into global coordinated constraints. Under the influence of complex and drastically fluctuating frequency regulation commands, when the supercapacitor's capacity is depleted or its power limit is reached, high-frequency spike power commands overflow to the lithium battery or thermal power unit side, forcing the control valve to oscillate at high frequency within a very small opening range, causing valve core wear and hydraulic actuator jamming. Simultaneously, under the high-frequency shallow charge and discharge micro-cycle, the solid electrolyte interface film of the lithium battery ruptures rapidly, resulting in an actual lifespan far below design expectations. More critically, there is a nonlinear cross-coupling and deterioration relationship between damper fatigue and battery degradation. Battery degradation leads to accelerated damper fatigue due to high-frequency command feedback, while damper limitation transfers low- and medium-frequency fluctuations to the energy storage system, worsening battery degradation. Both exhibit synergistic collapse characteristics under dual sub-health conditions. Current technologies lack an evaluation system that can uniformly quantify mechanical fatigue and electrochemical degradation and dynamically capture their coupling relationship, resulting in the inability to find adaptive optimal parameter boundaries for filter cutoff frequency and power distribution dead zone between protecting the damper and protecting the battery.

[0004] Therefore, an optimized frequency regulation scheme for hybrid energy storage thermal power units is desired. Summary of the Invention

[0005] To address the aforementioned technical problems, this application is proposed. Embodiments of this application provide a frequency regulation method for hybrid energy storage thermal power units based on hierarchical adaptive control.

[0006] According to one aspect of this application, a frequency regulation method for hybrid energy storage thermal power units based on hierarchical adaptive control is provided, comprising:

[0007] S1: Extract feature sequences from the real-time collected valve travel data, battery operation data, supercapacitor charge data, and frequency modulation commands to obtain travel feature sequences, discharge depth sequences, charge state sets, and preprocessed frequency modulation commands;

[0008] S2: Mechanical fatigue damage evolution and microcirculation equivalent capacity loss quantification are performed on the travel characteristic sequence and discharge depth sequence to obtain fatigue index and attenuation index;

[0009] S3: Construct a two-dimensional health status coordinate system based on the fatigue index and the decay index, and use a multi-input multi-output fuzzy logic reasoning engine to perform potential field optimization deduction on the two-dimensional health status coordinate system to obtain the adjustment coefficient vector.

[0010] S4: Based on the frequency band offset weight and dead zone amplification weight in the adjustment coefficient vector, adaptive frequency conversion and dead zone parameter update are performed on the fundamental frequency and power allocation dead zone bandwidth of the high-pass and low-pass filters combined with the state of charge set to obtain the cutoff frequency parameter and dead zone width parameter.

[0011] S5: High-frequency glitches are truncated and filtered in the preprocessed frequency modulation command using the dead zone width parameter to obtain the effective frequency modulation command. The effective frequency modulation command is then decoupled in the hierarchical frequency domain based on the cutoff frequency parameter to obtain the thermal power command, battery power command and supercapacitor power command.

[0012] Further, step S1 includes: S1.1: Resample and align the valve travel data, battery operation data, and supercapacitor charge data to obtain aligned valve travel data, aligned battery status data, and aligned supercapacitor charge data; perform smoothing filtering on the frequency modulation command to obtain a preprocessed frequency modulation command. S1.2: Extract stroke feature sequences from the aligned valve stroke data; S1.3: Unpack the aligned battery state data to separate the current sub-item and the state of charge sub-item, and perform discharge depth throughput conversion on the current sub-item to obtain the discharge depth sequence; S1.4: Perform dimensional concatenation and vectorization on the aligned battery charge data and the aligned supercapacitor charge data to obtain the charge state set.

[0013] Further, in S1.2, a stroke feature sequence is extracted from the aligned valve stroke data using the following formula: ; in, For the first The instantaneous percentage of the aligned turbine control valve stroke opening data input at any given time; For the first The instantaneous percentage of the aligned turbine control valve stroke opening data input at any given time. For the first The cumulative travel value in the cumulative travel feature sequence of the output at any time.

[0014] Furthermore, S1.3 includes: calculating the discharge depth throughput of the current sub-term using the following formula: ; in, For the first The instantaneous current value of the lithium battery obtained from constant unpacking. The discrete sampling time step of the system, This refers to the rated nominal capacity of the lithium battery system.

[0015] Further, step S2 includes: S2.1: Input the travel characteristic sequence into the rainflow counting method fatigue life model to obtain the fatigue index; S2.2: Input the discharge depth sequence into the electrochemical Arrhenius capacity decay model to obtain the decay index.

[0016] Further, step S3 includes: S3.1: Based on the preset two-dimensional Euclidean space, the fatigue index is used as the first dimension and the decay index is used as the second dimension to perform orthogonal combination and vectorization mapping to obtain two-dimensional health status coordinates. S3.2: Input the two-dimensional health status coordinates into the multi-input multi-output fuzzy logic inference engine to activate the frequency band offset weight and dead zone amplification weight respectively to perform potential field optimization inference to obtain the adjustment coefficient vector.

[0017] Further, step S4 includes: S4.1: Extract the frequency band offset weight in the adjustment coefficient vector, and combine it with the remaining capacity margin of each energy storage unit in the state of charge set that deviates from the ideal center value as a multiplication operator to perform frequency conversion boundary sliding based on health status weight on the preset high and low pass filter base frequency to obtain the cutoff frequency parameter. S4.2: Extract the dead zone amplification weight from the adjustment coefficient vector, and combine it with the capacity extreme value of the energy storage unit with the largest deviation of the state of charge concentration, to stretch and expand the preset basic dead zone bandwidth to obtain the dead zone width parameter.

[0018] Further, step S5 includes: S5.1: Based on the dead-time width parameter, the preprocessed frequency modulation command is truncated and effective features are extracted under controlled dead-time conditions to obtain an effective frequency modulation command; S5.2: Based on the low-pass cutoff frequency component in the cutoff frequency parameter, the effective frequency modulation command is subjected to low-pass frequency conversion filtering and decoupled from the thermal power reference power to obtain the thermal power power command and the medium- and high-frequency residual power command. S5.3: Based on the high-pass cutoff frequency component in the cutoff frequency parameter, perform ultra-high frequency pulse stripping on the mid-to-high frequency residual power command to obtain the supercapacitor power command, and subtract the supercapacitor power command from the mid-to-high frequency residual power command to obtain the lithium battery power command.

[0019] Further, step S3.2 includes: S3.2.1: Construct a coupled penalty tensor based on the fatigue index and decay index in the two-dimensional health status coordinates; S3.2.2: The collaborative crisis index is obtained by using the coupling penalty tensor to measure the collaborative crisis index of the two-dimensional health status coordinates; S3.2.3: Based on the preset crisis tolerance threshold, the collaborative crisis index is shifted and saturated to obtain the adjustment coefficient vector.

[0020] Compared with existing technologies, this application provides a frequency regulation method for hybrid energy storage thermal power units based on hierarchical adaptive control. It extracts feature sequences from real-time collected valve travel data, battery operation data, supercapacitor charge data, and frequency regulation commands. Based on these, it performs mechanical fatigue wear evolution on the accumulated valve travel and quantifies the micro-cycle equivalent capacity loss of the battery discharge depth to obtain fatigue and attenuation indices. These indices are then mapped to two-dimensional health status coordinates and used by a fuzzy logic inference engine for potential field optimization to generate a regulation coefficient vector. This vector, combined with the energy storage charge state, adaptively updates the filter cutoff frequency and power allocation dead zone. Finally, based on the updated parameters, it performs dead zone truncation filtering and hierarchical frequency domain decoupling on the frequency regulation commands, outputting power commands for the thermal power, battery, and supercapacitor respectively. This achieves global coordinated protection of valve wear and battery life while ensuring frequency regulation performance. Attached Figure Description

[0021] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0022] Figure 1 This is a flowchart of a hybrid energy storage thermal power unit frequency regulation method based on hierarchical adaptive control according to an embodiment of this application;

[0023] Figure 2 This is a schematic diagram of the data flow of the frequency regulation method for hybrid energy storage thermal power units based on hierarchical adaptive control according to an embodiment of this application;

[0024] Figure 3 This is a flowchart of step S1 of the frequency regulation method for hybrid energy storage thermal power units based on hierarchical adaptive control according to an embodiment of this application;

[0025] Figure 4 This is a flowchart of step S3 of the frequency regulation method for hybrid energy storage thermal power units based on hierarchical adaptive control according to an embodiment of this application;

[0026] Figure 5 This is a flowchart of step S5 of the frequency regulation method for hybrid energy storage thermal power units based on hierarchical adaptive control according to an embodiment of this application. Detailed Implementation

[0027] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0028] In the actual operation of frequency regulation of hybrid energy storage and thermal power, the existing hierarchical control scheme only adjusts the parameters based on the state of charge of energy storage, without incorporating the mechanical fatigue of the regulating valve and the electrochemical degradation of the battery into the decision-making framework. This results in the overflow of high-frequency glitch commands, causing ineffective oscillation and wear of the regulating valve and a sharp drop in the micro-cycle life of the battery. Furthermore, the deterioration of the cross-coupling relationship between the two cannot be effectively perceived and suppressed. To address the aforementioned issues, this solution performs timestamp alignment and feature sequence extraction on the valve travel, battery operating status, supercapacitor state of charge, and frequency modulation commands at the data acquisition layer, obtaining travel feature sequences, discharge depth sequences, state of charge sets, and preprocessed frequency modulation commands after smoothing and filtering. At the health assessment layer, the travel feature sequences are input into a rainflow counting fatigue life model to quantify the mechanical damage to the valve, and the discharge depth sequences are input into an electrochemical Arrhenius capacity decay model to quantify the micro-cycle loss of the battery, yielding fatigue and decay indices respectively. At the decision generation layer, the fatigue and decay indices are orthogonally combined into a two-dimensional health status coordinate system, which is then fed into a multi-input multi-output fuzzy logic inference engine for potential field optimization, outputting an adjustment coefficient vector containing frequency band offset weights and dead zone amplification weights. At the parameter update layer, the frequency band offset weights are used... By combining the remaining capacity margin of each energy storage unit in the state of charge concentration (SCC) deviating from the ideal center value, the fundamental frequency of the high-pass and low-pass filters is subjected to frequency conversion boundary sliding. The dead-zone bandwidth is expanded and contracted by using dead-zone amplification weights and the extreme value of the capacity of the energy storage unit with the largest deviation in SCC, resulting in adaptively updated cutoff frequency parameters and dead-zone width parameters. At the instruction execution layer, the preprocessed frequency modulation instructions are first filtered for high-frequency glitches using the dead-zone width parameter to obtain effective frequency modulation instructions. Then, based on the cutoff frequency parameter, the effective frequency modulation instructions are sequentially subjected to low-pass frequency conversion filtering to decouple the thermal power instruction and the medium-to-high frequency residual power instruction. The medium-to-high frequency residual power instruction is then stripped by ultra-high frequency pulses to separate the supercapacitor power instruction and the lithium battery power instruction. This allows the power allocation boundary to dynamically migrate with the health status of the equipment, achieving global collaborative optimization between frequency modulation performance and equipment life protection. Figure 1 This is a flowchart of a hybrid energy storage thermal power unit frequency regulation method based on hierarchical adaptive control according to an embodiment of this application. Figure 2 The data flow diagram of the frequency regulation method for hybrid energy storage thermal power units based on hierarchical adaptive control according to the embodiments of this application is shown below. Figure 1 and Figure 2As shown, the frequency regulation method for hybrid energy storage thermal power units based on hierarchical adaptive control according to an embodiment of this application includes: S1, extracting feature sequences from real-time collected valve travel data, battery operation data, supercapacitor charge data, and frequency regulation commands to obtain travel feature sequences, discharge depth sequences, charge state sets, and preprocessed frequency regulation commands; S2, quantifying mechanical fatigue damage evolution and micro-circulation equivalent capacity loss based on the travel feature sequences and discharge depth sequences to obtain fatigue index and attenuation index; S3, constructing a two-dimensional health status coordinate system based on the fatigue index and attenuation index, and utilizing multi-input multi-output fuzzy logic... The logical inference engine performs potential field optimization deduction on the two-dimensional health state coordinates to obtain the adjustment coefficient vector; S4, based on the frequency band offset weight and dead zone amplification weight in the adjustment coefficient vector, adaptive frequency conversion and dead zone parameter updates are performed on the fundamental frequency of the high-pass and low-pass filters combined with the power allocation dead zone bandwidth to obtain the cutoff frequency parameter and dead zone width parameter; S5, the dead zone width parameter is used to perform high-frequency glitch truncation filtering on the preprocessed frequency modulation command to obtain the effective frequency modulation command, and the effective frequency modulation command is decoupled into a layered frequency domain based on the cutoff frequency parameter to obtain the thermal power power command, battery power command and supercapacitor power command.

[0029] Specifically, in step S1, feature sequences are extracted from the real-time collected valve travel data, battery operation data, supercapacitor charge data, and frequency modulation commands to obtain travel feature sequences, discharge depth sequences, charge state sets, and preprocessed frequency modulation commands. It should be understood that because the valve travel sensor, lithium battery management system, and supercapacitor monitoring module of the thermal power unit operate independently in the industrial field, their data acquisition frequencies and time bases have inherent differences. Furthermore, the frequency modulation commands issued by the power grid dispatch center contain high-frequency white noise interference. Directly inputting these asynchronously sampled, noisy raw data into the subsequent health status assessment and power allocation stages will lead to statistical distortion of the valve travel mileage, calculation deviations in battery micro-circulation throughput, and timing errors in control decisions. Therefore, in the technical solution of this application, feature sequences are extracted from the real-time collected valve travel data, battery operation data, supercapacitor charge data, and frequency modulation commands to obtain travel feature sequences, discharge depth sequences, charge state sets, and preprocessed frequency modulation commands. This transforms the multi-source heterogeneous raw operation data into a standardized feature dataset that is time-aligned, semantically clear, and directly usable for subsequent fatigue quantification and adaptive parameter updates. This eliminates timestamp discrepancies and command noise interference between multi-source data, providing a reliable data foundation for the accurate calculation of subsequent fatigue and attenuation indices. Simultaneously, the charge state information of the lithium battery and supercapacitor is reconstructed into a unified vectorized representation, ensuring the integrity and consistency of the data flow in the hierarchical adaptive control link.

[0030] Figure 3This is a flowchart of step S1 of the frequency regulation method for hybrid energy storage thermal power units based on hierarchical adaptive control according to an embodiment of this application. Figure 3 As shown, step S1 includes: S1.1, resampling and aligning the valve travel data, battery operation data, and supercapacitor charge data to obtain aligned valve travel data, aligned battery state data, and aligned supercapacitor charge data, and smoothing and filtering the frequency modulation command to obtain a preprocessed frequency modulation command; S1.2, extracting travel feature sequences from the aligned valve travel data; S1.3, unpacking the aligned battery state data to separate the current sub-item and the state of charge sub-item, and performing discharge depth throughput conversion on the current sub-item to obtain a discharge depth sequence; S1.4, performing dimensional concatenation and vectorization on the aligned battery charge data and the aligned supercapacitor charge data to obtain a state of charge set.

[0031] Accordingly, in step S1.1, the valve travel data, battery operation data and supercapacitor charge data are resampled and timestamped to obtain aligned valve travel data, aligned battery status data and aligned supercapacitor charge data, and the frequency modulation command is smoothed and filtered to obtain a preprocessed frequency modulation command. It is understandable that, since the turbine valve travel sensor, lithium battery management system, and supercapacitor monitoring module of a thermal power unit belong to different control subsystems in the industrial field, their data acquisition cycles and hardware clock references have inherent differences. The sampling frequency of the valve travel sensor is limited by the accuracy of mechanical displacement detection, the sampling frequency of the lithium battery management system depends on the update cycle of the electrochemical state estimation algorithm, and the sampling frequency of the supercapacitor monitoring module is determined by its rapid charging and discharging characteristics. The three data streams exhibit non-equal intervals and asynchronous discrete distribution characteristics on the time axis. If they are directly input into the subsequent travel feature sequence extraction and discharge depth throughput conversion stages without alignment processing, it will cause timing misalignment errors in the differential operation and cumulative integration between adjacent sampling points, causing the quantitative results of the valve cumulative travel and battery micro-circulation throughput to deviate from the actual physical process. At the same time, the frequency modulation command issued by the power grid dispatch center inevitably has high-frequency white noise and communication spikes superimposed in the transmission link. If these interference components that are not intended for frequency modulation are not filtered out, they will cause false triggering or missed triggering in the subsequent dead zone truncation determination, interfering with the accuracy of power allocation decisions. Therefore, in the technical solution of this application, the valve travel data, battery operation data, and supercapacitor charge data are resampled and timestamped to obtain aligned valve travel data, aligned battery status data, and aligned supercapacitor charge data. The frequency modulation command is then smoothed and filtered to obtain a preprocessed frequency modulation command. This eliminates time base deviations between multi-source heterogeneous data and suppresses high-frequency noise interference in the frequency modulation command. This ensures that the input data upon which subsequent travel feature sequence extraction and discharge depth sequence calculation depend are strictly point-to-point in the time dimension, avoiding differential calculation distortion and accumulated quantization deviation caused by asynchronous sampling. Simultaneously, the preprocessed frequency modulation command retains only the effective power fluctuation component reflecting the actual frequency modulation demand of the power grid, providing a clean command signal basis for subsequent dead-zone width parameter truncation filtering and layered frequency domain decoupling.

[0032] Specifically, in a specific example of this application, a unified discrete sampling time step is first determined for the system. Using this step as the reference time grid, resampling timestamp alignment operations are performed on the valve travel data, battery operation data, and supercapacitor charging data, respectively. That is, for each channel of raw data, the nearest preceding raw sample value is found at each node of the unified time grid, and the zero-order hold method is used to fill the current grid node with the preceding sample value, so that the three channels of data form an equally spaced, point-by-point aligned sequence structure on the same discrete time axis, respectively obtaining aligned valve travel data, aligned battery status data, and aligned supercapacitor charging data. Next, smoothing filtering processing is performed on the frequency modulation command. That is, at each discrete sampling time, the original power amplitude of the frequency modulation command at the current time is combined with the power amplitude that has been smoothed at the previous time by a weighted linear combination. The weight distribution ratio between the current raw value and the historical smoothed value is controlled by a preset smoothing coefficient, so that the output signal retains the effective low-frequency adjustment components in the frequency modulation command while gradually attenuating the amplitude of high-frequency noise and communication glitches, thus obtaining the preprocessed frequency modulation command. In the aforementioned resampling alignment process, the zero-order hold method is selected based on the physical characteristic that the variation amplitude of the gate opening, battery current, and supercapacitor state of charge is limited within adjacent original sampling periods. This method achieves time axis unification with minimal computational overhead without introducing additional interpolation oscillations, and is suitable for real-time operation constraints of power plant distributed control systems. In the smoothing filtering process, the value of the smoothing coefficient determines the balance between the filter's suppression strength for high-frequency noise and the tracking delay of the effective frequency modulation signal. A smaller smoothing coefficient makes the filter rely more on historical smoothing values, thereby enhancing noise suppression capability but increasing signal tracking delay. A larger smoothing coefficient makes the filter follow the current original value more closely, thereby reducing delay but weakening noise suppression effect. In actual engineering deployment, the specific value of this coefficient is determined according to the typical noise spectrum characteristics of the power grid frequency modulation command and the allowable range of response delay for frequency modulation assessment.

[0033] Accordingly, in step S1.2, a stroke feature sequence is extracted from the aligned valve stroke data. It should be understood that, as the turbine valve is the direct mechanical actuator for executing frequency regulation commands in a thermal power unit, its lifespan does not depend on the absolute opening value at a given moment, but rather on the cumulative amount of reciprocating motion corresponding to the opening changes during continuous frequency regulation. Especially under high-frequency, small-amplitude, and frequent reversing frequency regulation conditions, the mechanical wear and fatigue impact borne by the valve core, linkage mechanism, and hydraulic actuator are mainly manifested as the continuous accumulation of displacement increments between adjacent sampling moments. If the aligned valve stroke data is directly input into the subsequent fatigue assessment stage as the original opening sequence, it is impossible to effectively distinguish between the static opening level and the dynamic motion intensity, thus failing to truly reflect the actual mechanical load of the valve in joint frequency regulation. Therefore, in the technical solution of this application, a stroke feature sequence is further extracted from the aligned valve stroke data to convert the original opening change process into a feature quantity that can characterize the cumulative motion mileage of the valve. This provides a direct, continuous, and quantifiable input basis for subsequent fatigue index calculations based on mechanical fatigue wear evolution. This allows the control system to no longer infer the valve load indirectly based solely on power response results, but to constrain subsequent power allocation strategies based on the valve's actual operating costs, thereby improving the accuracy of identifying and protecting against valve mechanical wear risks.

[0034] Specifically, in a specific example of this application, the valve travel data aligned with the resampled timestamps is first read. Then, using a system-wide unified discrete sampling time step, the instantaneous valve opening percentage corresponding to each sampling moment is extracted sequentially along the time axis to ensure that the opening data involved in the calculation are strictly continuous and comparable in the time dimension. Next, differential processing is performed on the instantaneous opening percentages of adjacent sampling moments to calculate the change in opening percentage relative to the previous sampling moment. The absolute value of this change is then taken to eliminate the mutual cancellation of the valve opening and closing directions in terms of sign, ensuring that each actual mechanical displacement is equivalently included in the motion consumption. Then, the absolute opening change at each sampling moment is accumulated and integrated along the discrete time axis to obtain a cumulative travel value that monotonically increases with time. This cumulative travel value is then used to construct a travel feature sequence, the calculation formula of which is:

[0035]

[0036] in, For the first The instantaneous percentage of the aligned turbine control valve stroke opening data input at any given time; For the first The instantaneous percentage of the aligned turbine control valve stroke opening data input at any given time. This represents the actual single opening displacement of the valve between two adjacent sampling times. For the first The cumulative travel value in the cumulative travel characteristic sequence of the output valve at any given time. Furthermore, the difference term in the above formula... This is used to extract the dynamic response amplitude of the regulating gate to continuous frequency modulation commands. It reflects the actual strength of the gate's driven action within a single sampling period. The significance of absolute value calculation lies in treating opening and closing actions as a unified contribution to mechanical wear, avoiding cancellation due to opposite action directions in mathematical summation, thus fully preserving the actual mechanical displacement cost of the regulating gate. The significance of the cumulative summation part lies in integrating discrete small-amplitude high-frequency actions into a continuously tracked total action mileage, allowing frequency modulation disturbances with small individual amplitudes but extremely high frequency to fully manifest their erosive effect on the regulating gate's lifespan after accumulation. Furthermore, after outputting the obtained stroke characteristic sequence to the subsequent mechanical fatigue wear evolution stage, the subsequent stages can evaluate the regulating gate's fatigue state based on this stroke characteristic sequence rather than the original opening sequence. This enables the joint frequency modulation control strategy to identify the additional mechanical burden caused by high-frequency residual power to the regulating gate and provides a state basis directly related to equipment lifespan for subsequent adjustments to cutoff frequency parameters and dead zone width parameters.

[0037] Accordingly, in step S1.3, the aligned battery state data is unpacked to separate the current sub-item and the state of charge sub-item, and the discharge depth throughput is converted to the current sub-item to obtain the discharge depth sequence. It is understandable that the aligned battery state data is composite state data output by the battery management system, which simultaneously contains current information characterizing instantaneous charge and discharge behavior and state of charge information characterizing the remaining energy level. In the subsequent processing chain of this application, these two types of information perform different functions. The current information is used to quantify the electrochemical throughput load of the lithium battery due to frequent power reversals during joint frequency regulation, and the state of charge information is used to construct the state of charge set together with the aligned supercapacitor charge data. If the aligned battery state data is not unpacked, the subsequent degradation assessment and capacity boundary judgment will not obtain semantically clear and physically single input data. At the same time, the life loss of lithium battery is not determined by the current direction at a certain moment, but by the cumulative charge and discharge throughput during continuous frequency regulation operation. Especially under the frequency regulation conditions of high frequency, small amplitude, and repeated reversal, although the amplitude of a single current is limited, the long-term accumulated ampere-hour throughput will significantly accelerate the loss of active materials and interface side reactions. If only the instantaneous current value is used as the input of the subsequent electrochemical degradation model, it cannot truly reflect the micro-cycle aging intensity of the lithium battery in the joint frequency regulation scenario. Therefore, in the technical solution of this application, the aligned battery state data is further unpacked to separate the current sub-item and the state of charge sub-item, and the discharge depth throughput of the current sub-item is converted to obtain the discharge depth sequence. This decomposes the composite state data into dedicated data objects for life assessment and capacity scheduling, respectively, and transforms the discrete instantaneous current response into a standardized feature sequence that can be continuously accumulated and directly characterize the micro-cycle load. In this way, the subsequent electrochemical Arrhenius capacity decay model can receive input consistent with the battery aging mechanism, thereby improving the authenticity and stability of the decay index calculation. At the same time, it retains an independent state of charge sub-item for the subsequent construction of the state of charge set, ensuring that the data links of the energy storage system capacity boundary adjustment and life assessment are independent and have a complete closed loop.

[0038] Specifically, in one example of this application, the battery state data, after being resampled and timestamp aligned, is first read and parsed at the field level according to the data frame structure preset by the battery management system. The data frame corresponding to the same sampling time is split into a current sub-item and a state of charge sub-item. The current sub-item is used to characterize the instantaneous charge and discharge intensity of the lithium battery in the current sampling period, and the state of charge sub-item is used to characterize the remaining capacity level of the lithium battery at the end of the current sampling period. Then, the parsed current sub-items are arranged in the order of sampling time, and the instantaneous current value of the lithium battery at each sampling time is extracted. The absolute value of the instantaneous current value is then processed to eliminate the mutual cancellation effect of the charging current and discharging current in terms of sign, so that the entire electrochemical throughput of the battery during the bidirectional power regulation process is included in the lifetime consumption. Then, the absolute value of the current at each sampling moment is multiplied by the discrete sampling time step of the system to obtain the power throughput increment within that sampling period. This increment is then divided by the rated nominal capacity of the lithium battery system. The throughput under different time lengths and capacity levels is uniformly converted into a dimensionless depth of discharge increment relative to the nominal capacity. Finally, the depth of discharge increments for each sampling period are accumulated along the time axis to form a depth of discharge sequence. The calculation formula is as follows:

[0039] The depth of discharge throughput of the current sub-term is calculated using the following formula:

[0040]

[0041] in, This represents the cumulative discharge depth value in the discharge depth sequence output at the k-th sampling time. For the first The instantaneous current value of the lithium battery obtained from constant unpacking. This represents the current throughput amplitude obtained by taking the absolute value of the instantaneous current of the lithium battery. This represents the discrete sampling time step of the system. This represents the rated nominal capacity of the lithium battery system. The constant 3600 is used to convert the sampling time step, measured in seconds, into a capacity unit measured in hours. Furthermore, the absolute value term in the above formula... In this scenario, the bidirectional electrochemical stress generated by the combined frequency modulation of lithium batteries is uniformly characterized. Regardless of whether the current moment represents the power absorbed during charging or the power released during discharging, both factors contribute to the aging of the battery's internal active materials and interface structure, and are therefore considered as contributions to lifespan degradation. (Product term) The significance of this scenario lies in converting the instantaneous current response into the actual power throughput within a single sampling period, thereby mapping the short-term power fluctuations caused by grid frequency regulation into the actual electrochemical exchange load experienced inside the battery; denominator part The throughput is normalized based on the rated nominal capacity of the lithium battery system, so that the discharge depth sequence no longer depends on the specific battery size, but directly represents the relative strength of the current operating strategy's impact on battery life. The significance of the cumulative summation part is to integrate a large number of discrete, small-amplitude bidirectional power exchanges under high-frequency modulation conditions into a continuously increasing cumulative loss, thereby accurately revealing the long-term erosion process of lithium battery life caused by microcirculation effects. Furthermore, the state of charge sub-item separated during the data unpacking process is retained for subsequent dimensional concatenation and vectorization processing with the aligned supercapacitor charge data, while the discharge depth sequence is output to the subsequent electrochemical Arrhenius capacity decay model, so that the subsequent decay index calculation has a clear scenario source, continuous time accumulation characteristics, and an input form that matches the battery aging mechanism.

[0042] Accordingly, in step S1.4, the aligned battery charge data and the aligned supercapacitor charge data are dimensionally concatenated and vectorized to obtain a charge state set. It should be understood that, due to the fundamental differences in the physical functions undertaken by lithium batteries and supercapacitors in a hybrid energy storage and thermal power joint frequency regulation system—lithium batteries correspond to medium-frequency energy regulation tasks, while supercapacitors correspond to ultra-high-frequency power regulation tasks—although both belong to energy storage units, their remaining capacity margin, sustainable power output capability, and constraints on subsequent cutoff frequency parameters and dead zone width parameters are different. If only the aligned battery charge data and the aligned supercapacitor charge data are retained as two independent scalar data streams, the subsequent adaptive parameter update stage will find it difficult to synchronously perceive the joint capacity state of the two types of energy storage units within the same computational framework, and it will also be impossible to directly complete a unified comparison and collaborative judgment of the degree of deviation from the ideal center value. Therefore, in the technical solution of this application, the aligned battery charge data and the aligned supercapacitor charge data are further dimensionally concatenated and vectorized to obtain a charge state set. This reconstructs the dispersed single energy storage charge state information into a joint state data object that can comprehensively characterize the real-time capacity boundary of the hybrid energy storage system. This allows the subsequent adjustment coefficient vectors, when applied to the cutoff frequency parameter and dead zone width parameter, to directly utilize the charge state set for unified invocation, synchronous comparison, and coordinated constraint of the remaining capacity margin of each energy storage unit. This avoids the power distribution imbalance caused by local adjustments based solely on the state of a single energy storage unit and improves the capacity utilization balance and control decision consistency of the hybrid energy storage system during joint frequency regulation.

[0043] Specifically, in a specific example of this application, the battery charge data and the aligned supercapacitor charge data, after being resampled and timestamped, are first read and synchronously matched at a unified discrete sampling time to ensure that the two charge state data streams involved in the processing correspond to the energy storage operation state at the same physical moment. Next, the aligned battery charge data and the aligned supercapacitor charge data are consistently organized to confirm that both use a unified charge state expression scale, making the subsequently formed joint state data directly comparable and having unified engineering interpretation significance. Then, according to a preset dimensional order, the aligned battery charge data is concatenated as a one-dimensional state component and the aligned supercapacitor charge data as another-dimensional state component, and a charge state set is generated in a vectorized form, thereby transforming the original separate charge state data into a unified energy storage operation state. The two separately stored state-of-charge variables are integrated into a unified state object that can be directly called by subsequent control links. Furthermore, this state-of-charge set is directly used in the subsequent adaptive frequency conversion and dead-zone parameter update links to characterize the remaining capacity margin of each energy storage unit deviating from the ideal center value. The state component corresponding to the lithium battery is used to reflect its sustainable energy support capability when undertaking the task of medium-frequency power compensation, and the state component corresponding to the supercapacitor is used to reflect its transient capacity margin when undertaking the task of high-frequency pulse absorption. Based on this, the control system performs frequency conversion boundary sliding on the basic frequency of the high-pass and low-pass filters and adjusts the basic dead-zone bandwidth, thereby avoiding mechanically maintaining the predetermined power allocation strategy when the capacity of a certain energy storage unit approaches the boundary, and finally realizing a more coordinated joint frequency modulation control among thermal power units, lithium batteries and supercapacitors.

[0044] Specifically, in step S2, the travel characteristic sequence and discharge depth sequence are subjected to mechanical fatigue damage evolution and micro-circulation equivalent capacity loss quantification to obtain fatigue index and degradation index. It should be understood that although the travel characteristic sequence and discharge depth sequence have quantified the cumulative operating mileage of the control valve and the cumulative electrochemical throughput of the lithium battery, these two types of characteristic sequences only reflect the physical load intensity experienced by the equipment during joint frequency regulation, and have not yet been transformed into normalized evaluation indicators that can directly characterize the current degree of equipment health degradation. Furthermore, the subsequent construction of the two-dimensional health status coordinates and the generation of the adjustment coefficient vector both require input of health status quantities with unified dimensions that can be directly compared across dimensions. If the travel characteristic sequence and discharge depth sequence are directly input into the subsequent decision-making process, the control system cannot determine the proportion of the current cumulative load relative to the equipment's lifespan limit, nor can it establish a comparable risk measurement benchmark between the mechanical fatigue of the control valve and the electrochemical degradation of the battery. Therefore, in the technical solution of this application, mechanical fatigue damage evolution and micro-circulation equivalent capacity loss quantification are further performed on the travel characteristic sequence and discharge depth sequence to obtain fatigue index and attenuation index, thereby transforming the cumulative load characteristics into a normalized health degradation index that can characterize the proportion of equipment lifespan consumption. This allows the subsequent construction of a two-dimensional health state coordinate system to obtain two health state inputs that physically correspond to the mechanical and electrochemical domains respectively and are directly comparable on a numerical scale. This provides an accurate risk measurement basis for the potential field optimization deduction of the adjustment coefficient vector, enabling the control system to make power allocation decisions based on the actual lifespan consumption of the equipment rather than solely on the energy storage state of charge.

[0045] More specifically, in the embodiments of this application, step S2 includes: S2.1: inputting the travel characteristic sequence into the rainflow counting method fatigue life model to obtain the fatigue index; S2.2: inputting the discharge depth sequence into the electrochemical Arrhenius capacity decay model to obtain the decay index.

[0046] Specifically, in a particular example of this application, the travel characteristic sequence is first input into the rainflow counting method fatigue life model for mechanical fatigue damage evolution calculation. This model is based on the power-law damage accumulation theory in material fatigue mechanics. It calculates the ratio of the accumulated travel distance of the valve during the joint frequency modulation process to the total life limit travel distance of the valve material, and applies a nonlinear power function to amplify this ratio through a material fatigue sensitivity index. This maps the linearly increasing accumulated travel distance to the proportion of fatigue damage with accelerated deterioration characteristics, resulting in the fatigue index, the calculation formula of which is:

[0047]

[0048] in, This represents the fatigue index output at the k-th sampling time. This represents the cumulative travel value in the travel feature sequence input at the k-th sampling time. This represents the preset lifespan limit stroke constant of the turbine control valve material. The material sensitivity index represents the evolution of mechanical fatigue. In the above formula, the ratio term... The accumulated travel distance of the regulating valve during current frequency regulation operation is converted into a normalized proportion relative to its full life limit. This makes the fatigue index no longer dependent on the absolute travel value of a specific regulating valve model, but directly reflects the degree of erosion of the remaining life of the regulating valve by the current operating strategy. The power function operation can characterize the nonlinear cumulative characteristics of fatigue damage in metal materials. That is, when the accumulated travel distance of the regulating valve approaches its full life limit, the growth rate of the fatigue index increases significantly. This is consistent with the physical law of accelerated crack propagation in the regulating valve core and linkage mechanism in the later stage of fatigue, enabling the control system to obtain a more sensitive risk signal when the regulating valve enters the fatigue acceleration period. Next, the discharge depth sequence is input into the electrochemical Arrhenius capacity decay model for the quantification calculation of micro-circulation equivalent capacity loss. This model is based on the Arrhenius reaction rate theory in electrochemical kinetics. It combines the accumulated discharge depth throughput of the lithium battery during joint frequency regulation with the temperature dependence of the electrochemical side reactions inside the battery. Through the composite operation of nonlinear power function and exponential decay function, the accumulated throughput is mapped to the equivalent capacity loss proportion to obtain the decay index. The calculation formula is as follows:

[0049]

[0050] in, This represents the attenuation index of the output at the k-th sampling time. This represents the cumulative discharge depth value in the discharge depth sequence input at the k-th sampling time. This represents the preset electrochemical pre-degradation comprehensive factor constant. The power-law constant represents the nonlinear dependence of capacity loss on throughput. The apparent activation energy represents the electrochemical reaction inside a lithium battery. Represents the ideal gas constant. This represents the real-time absolute operating temperature of the lithium battery cell. In the above formula, the power function term... This is used to characterize the nonlinear response of lithium battery capacity decay to microcycle throughput. Specifically, as the cumulative depth of discharge continues to increase, the repeated rupture and reconstruction of the solid electrolyte interface film and the irreversible consumption of active lithium gradually accelerate the capacity loss rate, using a power-law approach. The steepness of this accelerated degradation process was controlled; the exponential decay term... This involves introducing the modulation effect of battery operating temperature on the rate of electrochemical side reactions. During joint frequency modulation, the temperature rise of the lithium battery due to continuous high-frequency charging and discharging lowers the activation energy barrier and accelerates the side reaction process. This index term allows the degradation index to be dynamically adjusted according to changes in the actual battery operating temperature, thus more realistically reflecting the actual aging rate of the battery under current thermal conditions. The significance of the pre-factor μμ lies in comprehensively absorbing the influence of inherent properties such as battery material system, electrolyte formulation, and manufacturing process on the degradation baseline, enabling the model to adapt to different types of lithium battery systems. Then, the calculated fatigue index and degradation index are simultaneously output to the subsequent two-dimensional health status coordinate construction stage, allowing the subsequent stages to perform orthogonal combination and collaborative risk assessment in a unified two-dimensional space based on these two normalized indices that respectively characterize the degree of life consumption in the mechanical and electrochemical domains.

[0051] Specifically, in step S3, a two-dimensional health status coordinate system is constructed based on the fatigue index and the decay index. A multi-input multi-output fuzzy logic inference engine is then used to perform potential field optimization on the two-dimensional health status coordinate system to obtain the adjustment coefficient vector. It should be understood that since the fatigue index and decay index quantify the lifespan consumption of the control valve and lithium battery from the mechanical and electrochemical domains respectively, but they still exist as independent scalars in the preceding steps, the subsequent adaptive frequency conversion and dead-zone parameter update stages need to simultaneously perceive the health degradation status of both types of equipment and generate parameter adjustment weights with clear control significance. If the fatigue index and decay index are not integrated into a unified state space for collaborative evaluation, the control system cannot simultaneously weigh the priority relationship between the control valve protection requirement and the battery protection requirement in the same decision-making process, nor can it automatically determine the adjustment direction and adjustment magnitude of the cutoff frequency parameter and dead-zone width parameter based on the joint health status of the two types of equipment. Therefore, in the technical solution of this application, a two-dimensional health status coordinate system is further constructed based on the fatigue index and the attenuation index. A multi-input multi-output fuzzy logic inference engine is then used to perform potential field optimization and deduction on the two-dimensional health status coordinate system to obtain the adjustment coefficient vector. This integrates the two independent health degradation indicators into a unified two-dimensional status representation. Furthermore, fuzzy inference transforms this status representation into control weights that can directly drive subsequent parameter updates. This allows the adaptive adjustment of subsequent cutoff frequency parameters and dead zone width parameters to no longer rely on a single-dimensional state judgment, but rather to make global collaborative decisions based on the combined status of tuning mechanical fatigue and battery electrochemical attenuation. This achieves synchronous protection of two types of critical equipment while ensuring tuning performance.

[0052] Figure 4 This is a flowchart of step S3 of the frequency regulation method for hybrid energy storage thermal power units based on hierarchical adaptive control according to an embodiment of this application. Figure 4As shown, step S3 includes: S3.1: Based on the preset two-dimensional Euclidean space, the fatigue index as the first dimension and the attenuation index as the second dimension are orthogonally combined and vectorized to obtain two-dimensional health status coordinates; S3.2: The two-dimensional health status coordinates are input into the multi-input multi-output fuzzy logic inference engine to excite the frequency band offset weight and dead zone amplification weight respectively to perform potential field optimization inference to obtain the adjustment coefficient vector.

[0053] Accordingly, in step S3.1, based on a preset two-dimensional Euclidean space, the fatigue index as the first dimension and the decay index as the second dimension are orthogonally combined and vectorized to obtain two-dimensional health status coordinates. It should be understood that since the fatigue index and decay index are independently calculated and generated by the rainflow counting fatigue life model and the electrochemical Arrhenius capacity decay model respectively in the preceding steps, they exist in their respective calculation channels as unrelated scalars. However, the subsequent multi-input multi-output fuzzy logic inference engine needs to receive a unified input object that can simultaneously carry the mechanical fatigue state and the battery electrochemical decay state when performing potential field optimization inference, so as to collaboratively evaluate the joint health status of the two types of equipment in the same inference process. If the fatigue index and decay index are input into the inference engine as separate scalars, the engine cannot determine the relative positional relationship and comprehensive degradation degree between the two risk dimensions in a unified metric space, nor can it perform global optimization based on the spatial characteristics of the joint status. Therefore, in the technical solution of this application, a pre-defined two-dimensional Euclidean space is used as the basis. The fatigue index is used as the first dimension and the attenuation index is used as the second dimension. These are orthogonally combined and vectorized to obtain two-dimensional health status coordinates. This integrates two independent health degradation scalars into a joint state representation with clear spatial geometric meaning. In this way, the subsequent fuzzy logic inference engine can directly perceive the comprehensive health degradation direction and degree of the system based on the two-dimensional health status coordinates in a unified Euclidean space. This provides an input basis with spatial structure characteristics for potential field optimization deduction, so that the generated adjustment coefficient vector can simultaneously reflect the joint constraints of damping protection requirements and battery protection requirements.

[0054] Specifically, in a concrete example of this application, the fatigue index output by the rainflow counting fatigue life model and the decay index output by the electrochemical Arrhenius capacity decay model at the current sampling time are first read to confirm that both are dimensionless scalars that have undergone normalization, thus meeting the prerequisite for joint characterization on the same numerical scale. Then, using a pre-defined two-dimensional Euclidean space as the mapping basis, the fatigue index is designated as the first-dimensional coordinate component of this space, and the decay index is designated as the second-dimensional coordinate component. The two dimensions maintain an orthogonal and independent relationship; that is, changes in the fatigue index do not affect the coordinate values ​​of the decay index in its dimension, and vice versa. This ensures that the coordinate positions in the two-dimensional space can unambiguously reflect the independent contributions of the two types of health degradation states. Then, the two dimensional components are combined in a preset order to generate a two-dimensional health status coordinate in vectorized form. The position of this coordinate in two-dimensional Euclidean space has intuitive physical interpretation meaning. When both the fatigue index and the attenuation index approach zero, the coordinate point is located near the origin, indicating that both the valve and the battery are in a healthy state. When the fatigue index increases alone, the coordinate point shifts along the first dimension, indicating that the valve fatigue risk dominates. When the attenuation index increases alone, the coordinate point shifts along the second dimension, indicating that the battery attenuation risk dominates. When both increase simultaneously, the coordinate point moves away from the origin along the diagonal, indicating that the system faces a synergistic risk of dual degradation. Furthermore, the obtained two-dimensional health status coordinate is output to the subsequent multi-input multi-output fuzzy logic inference engine, enabling the inference engine to perform potential field optimization inference based on the real-time position characteristics of this coordinate in two-dimensional Euclidean space. According to the direction and distance of the coordinate deviating from the origin, the adjustment direction and adjustment magnitude of the frequency band offset weight and dead zone amplification weight are determined, thereby realizing joint decision-making for valve mechanical protection and battery life protection.

[0055] Accordingly, in step S3.2, the two-dimensional health status coordinates are input into the multi-input multi-output fuzzy logic inference engine to excite the frequency band offset weights and dead zone amplification weights for potential field optimization and deduction to obtain the adjustment coefficient vector. It should be understood that although the two-dimensional health status coordinates have integrated the mechanical fatigue of the control valve and the electrochemical decay of the battery into a unified two-dimensional Euclidean space, the coordinates themselves are only a static state description quantity and have not yet been transformed into control weights that can directly drive the subsequent cutoff frequency parameters and dead zone width parameters for adaptive updates. The subsequent frequency conversion boundary sliding and dead zone bandwidth scaling stages both require weight scalars with clear adjustment directions and amplitudes as input. If the two-dimensional health status coordinates are not further used for decision deduction, the control system cannot transform the spatial location information of the equipment's health status into parameter adjustment commands with engineering feasibility. Therefore, in the technical solution of this application, the two-dimensional health status coordinates are further input into a multi-input multi-output fuzzy logic inference engine to excite the frequency band offset weight and dead zone amplification weight respectively for potential field optimization inference to obtain the adjustment coefficient vector. This transforms the health status position information in two-dimensional space into control weights that can directly participate in subsequent parameter update calculations through nonlinear inference mapping. This enables the adaptive adjustment of subsequent cutoff frequency parameters and dead zone width parameters to obtain adjustment driving quantities that strictly correspond to the current joint health status of the equipment, making the adjustment amplitude monotonically increase with the degree of health status deterioration. This minimizes interference with frequency modulation performance when the equipment is healthy and applies timely protective parameter corrections when the equipment deteriorates.

[0056] Specifically, in a specific example of this application, the two-dimensional health status coordinates output from the previous step are first received and used as input variables for a multi-input multi-output fuzzy logic inference engine. The inference engine uses the fatigue index and attenuation index in the two-dimensional health status coordinates as two inputs and the frequency band offset weight and dead zone amplification weight as two outputs. Its internal potential field optimization inference process is based on a weighted Euclidean distance metric mechanism. That is, the weighted Euclidean distance from the two-dimensional health status coordinates to the origin of the state space is first calculated and used as a scalar metric to characterize the overall urgency of the system. The weighting matrix is ​​set as a static diagonal matrix. The two elements on the diagonal correspond to the sensitivity weights of the damping fatigue dimension and the battery attenuation dimension in the comprehensive evaluation, respectively. The off-diagonal elements are set to zero. That is, in this embodiment, damping fatigue and battery attenuation are regarded as two independent risk dimensions and weighted contributions are made separately. Next, based on the calculated weighted Euclidean distance, a nonlinear exponential mapping operation is applied to the frequency band offset weight and the dead zone amplification weight, respectively. This mapping adopts a saturated exponential function structure. When the weighted Euclidean distance approaches zero, i.e., the system is in a healthy state, the mapping output approaches zero, and the inference engine does not generate adjustment driving force. When the weighted Euclidean distance increases, i.e., the system's health deteriorates, the mapping output rapidly climbs along the exponential curve and gradually approaches the preset saturation gain upper limit, thus forming a nonlinear monotonically increasing relationship between the response strength of the adjustment weight and the degree of health deterioration. The calculation formula is as follows:

[0057]

[0058] in, Indicates the first The adjustment coefficient vector generated at the sampling time. This represents the frequency band offset weights in the vector used to drive the sliding of the cutoff frequency of subsequent filters. This represents the dead-zone amplification weight in the vector used to drive subsequent power allocation dead-zone scaling. Represents the transpose vector of the two-dimensional health status coordinates. This represents a predefined diagonal weighted matrix. The maximum saturation gain constant represents the frequency band offset weight. This represents the maximum saturation gain constant of the dead-zone amplification weight. This represents the repulsive field shape control parameter of the frequency regulation loop. This represents the shape control parameters of the repulsive field in the dead-zone adjustment loop. In the above formula, the quadratic operation... Can be used as a diagonal weighted matrix To calculate the squared weighted Euclidean distance from the two-dimensional health status coordinates to the origin of the state space for the metric benchmark, the diagonal weighting matrix is ​​used. The form is:

[0059]

[0060] in, The basic sensitivity coefficient of the self-weight represents the fatigue dimension of the tone. The base sensitivity weight coefficient for the battery degradation dimension is represented by the fact that off-diagonal elements are always zero, indicating that in this embodiment, no cross-coupling penalty term is introduced between the damping fatigue dimension and the battery degradation dimension. Each dimension contributes independently to the weighted distance according to its own sensitivity coefficient. Then, the arithmetic square root of the weighted Euclidean distance is multiplied by the shape control parameter. and And input to the saturated exponential mapping function. In this way, a nonlinear response curve can be constructed that starts from zero, monotonically increases, and has an upper bound saturation characteristic. When the health status coordinate is near the origin, the function output approaches zero, allowing the control system to apply no additional intervention. When the health status coordinate is far from the origin, the function output rapidly climbs to the saturation range, causing the control system to perform protective parameter adjustments with near-full amplitude. and The response sensitivity of the frequency regulation loop and dead-zone regulation loop to health deterioration is controlled separately. A larger shape control parameter makes the mapping curve steeper, thus triggering a stronger regulation response at lower levels of urgency, while a smaller shape control parameter makes the mapping curve flatter, thus improving tolerance to mild degradation; maximum saturation gain constant. and The significance of this scenario lies in limiting the physical upper limit of the frequency band offset weight and dead zone amplification weight, preventing the adjustment weight from increasing indefinitely under extreme and adverse conditions, which could cause the filter cutoff frequency or dead zone width to exceed the safe range allowed by the system hardware. Furthermore, the calculated frequency band offset weight and dead zone amplification weight are combined into an adjustment coefficient vector and output to the subsequent adaptive frequency conversion and dead zone parameter update stages.

[0061] Specifically, in industrial settings where thermal power units and hybrid energy storage systems jointly participate in automatic grid generation control and frequency regulation, the mechanical fatigue state of the turbine control valve and the electrochemical degradation state of the lithium battery are not two independent risk dimensions, but rather exhibit a deep-seated nonlinear cross-coupling and deterioration relationship. Specifically, when the lithium battery experiences severe capacity degradation and a significant decrease in response capability due to long-term micro-circulation, the underlying power distribution control system, when faced with high-frequency frequency regulation commands from the grid, is forced to overflow and transmit the high-frequency, irregularly spiked power commands that should have been handled by the energy storage system back to the thermal power unit. This forces the turbine control valve to perform high-frequency reciprocating oscillations within a very small opening range, thus rapidly accelerating the accumulation of mechanical fatigue in a short period. The reverse coupling path also holds true: when the control valve's movement amplitude or response rate is restricted by maintenance personnel due to severe mechanical wear, the low-to-medium frequency fluctuations that were originally smoothly absorbed by the thermal power unit are forced to be transferred to the energy storage system. The lithium battery then has to undertake additional charging and discharging tasks across a wider frequency band, further worsening its micro-circulation throughput and capacity degradation rate. More importantly, when both damper fatigue and battery degradation are in a sub-healthy state, the overall failure risk faced by the system is not a simple linear superposition of the two, but rather exhibits an exponentially amplified synergistic collapse characteristic.

[0062] However, in sub-step 3.2 of the first embodiment described above, the core mathematical model used to calculate the overall urgency level of the system is a weighted Euclidean distance, where the weighting matrix is ​​set to a static diagonal matrix, with off-diagonal elements always being zero. This mathematical modeling method fundamentally assumes that there is no statistical correlation or physical coupling between the mechanical fatigue dimension of the control valve and the electrochemical degradation dimension of the battery, directly negating the aforementioned cross-coupling synergistic penalty effect in the off-diagonal direction. In actual power plant operation, the consequence of this modeling defect is that when two critical pieces of equipment simultaneously enter the sub-health range, the control adjustment weights calculated by the algorithm are too weak and the response is lagging, failing to trigger sufficiently strong protective parameter adjustments in a timely manner, which may ultimately lead to irreversible safety accidents such as control valve jamming or battery thermal runaway.

[0063] To address the aforementioned deficiencies, a second embodiment is proposed. Specifically, in the second specific example of this application, step S3.2 includes: S3.2.1: constructing a coupling penalty tensor based on the fatigue index and decay index in the two-dimensional health status coordinates; S3.2.2: using the coupling penalty tensor to calculate the collaborative crisis index on the two-dimensional health status coordinates to obtain the collaborative crisis index; S3.2.3: performing translation bias and saturation mapping on the collaborative crisis index based on a preset crisis tolerance threshold to obtain an adjustment coefficient vector.

[0064] More specifically, in step S3.2.1, a coupled penalty tensor is constructed based on the fatigue index and decay index in the two-dimensional health status coordinates. It should be understood that, in the sub-steps of the first embodiment, the weighting matrix used to calculate the overall urgency of the system is set to a static diagonal matrix, with its off-diagonal elements always being zero. This mathematical modeling method fundamentally assumes that there is no statistical correlation or physical coupling between the mechanical fatigue dimension of the control valve and the electrochemical degradation dimension of the battery. However, in the actual operation of joint frequency regulation, the coupling strength between control valve fatigue and battery degradation is not constant, but dynamically changes with the degree of deterioration of their respective health states. When both are in the healthy range, the coupling effect is weak, while when both deteriorate simultaneously, the coupling effect is sharply enhanced. Specifically, when the lithium battery suffers severe capacity degradation and a significant decrease in response capability due to long-term micro-circulation, the underlying power distribution control system, when faced with high-frequency frequency regulation commands issued by the grid, has to overflow the high-frequency irregular glitch power commands that should have been handled by the energy storage system and send them back to the thermal power unit side, forcing the turbine control valve to perform high-frequency reciprocating within a very small opening range. The oscillating action rapidly accelerates the accumulation of mechanical fatigue in the control valve within a short period of time. The reverse coupling path also holds true. When the control valve's movement amplitude or response rate is restricted by maintenance personnel due to severe mechanical wear, the low- and medium-frequency fluctuation components that were originally smoothly absorbed by the thermal power unit will be forced to be transferred to the energy storage system. The lithium battery will have to undertake additional charging and discharging tasks in a wider frequency band, further deteriorating its micro-circulation throughput and capacity decay rate. More importantly, when the control valve fatigue and battery decay are both in a sub-healthy state, the comprehensive failure risk faced by the system is not a simple linear superposition of the two, but exhibits an exponentially amplified synergistic collapse characteristic. The static diagonal matrix modeling method in the first embodiment directly eliminates the aforementioned cross-coupling synergistic penalty effect in the off-diagonal direction. In actual power plant operation, the consequence of this modeling defect is that when two key devices simultaneously enter the sub-healthy range, the control and regulation weights calculated by the algorithm are weak and the response is lagging, failing to trigger a sufficiently strong protective parameter adjustment in time. Therefore, in the technical solution of this application, a coupled penalty tensor is further constructed based on the fatigue index and decay index in the two-dimensional health status coordinates. This allows for the precise capture of the dynamic coupling characteristics between valve mechanical fatigue and battery electrochemical decay as the operating conditions evolve at the mathematical model level. This overcomes the blind spot in modeling the coupling relationship between devices using the static diagonal matrix in the first embodiment. The overall criticality calculated by the system under dual sub-health conditions is significantly higher than the simple linear superposition of two single-dimensional risks. This provides a more advanced and sensitive risk warning signal for subsequent control decisions, effectively avoiding irreversible safety accidents such as valve jamming or battery thermal runaway caused by lag in control weight response.

[0065] In a specific example of this application, the fatigue index and decay index at the current sampling moment are first extracted from the two-dimensional health status coordinates, and used as the basic input variables for constructing the coupling penalty tensor. Next, the product of the fatigue index and the decay index is calculated, and a power function operation is performed on this product using a preset cross-coupling gain constant and a nonlinear co-excitation index to dynamically generate an off-diagonal covariance term. The magnitude of this covariance term adaptively expands with the degree of joint deterioration of the two health degradation indicators. Then, the generated off-diagonal covariance term is assembled with preset basic sensitivity weight coefficients for the damping fatigue dimension and battery decay dimension into a full-rank second-order symmetric tensor, resulting in the coupling penalty tensor, whose calculation formula is as follows:

[0066]

[0067] in, The coupling penalty tensor generated at the k-th sampling time. and These are the basic sensitivity weight coefficients for the damping fatigue dimension and the battery degradation dimension, respectively. The mechanical fatigue index of the valve is extracted from the two-dimensional health status coordinate matrix. The battery dynamic degradation index is extracted from the two-dimensional health status coordinate matrix. The cross-coupling gain constant is... This is a nonlinear co-excitation exponent, typically taking a value greater than 1. In the above formula, the diagonal elements... and The basic self-weight contributions of the damping fatigue dimension and the battery degradation dimension in the comprehensive urgency assessment are set respectively. Their function is consistent with the diagonal elements of the static diagonal weighted matrix in the first embodiment, ensuring that even under the condition of single-dimensional degradation, the health risk of that dimension can still make an independent contribution to the comprehensive distance measurement through its own sensitivity coefficient.

[0068] off-diagonal elements The physical meaning is as follows: When both the damper fatigue index and the battery degradation index approach zero (i.e., the equipment is healthy), their product is extremely small, the off-diagonal elements are approximately zero, and the tensor degenerates into a traditional diagonal matrix, without imposing additional coupling penalties. However, when both simultaneously rise into the sub-healthy range, the product term expands rapidly after being amplified by a power function, the off-diagonal elements increase significantly, and the determinant value of the tensor rises sharply, thus producing a synergistic penalty effect in subsequent distance metrics that far exceeds linear superposition. Among these, the product term... This function is used to capture the combined degradation of damping fatigue and battery decay. The product only makes a significant numerical contribution when both dimensions deteriorate simultaneously. If only one dimension deteriorates while the other remains healthy, the product approaches zero, off-diagonal elements automatically shrink, and the tensor naturally degenerates into a diagonal form. This avoids unnecessary cross-coupling interventions in single-device degradation scenarios. (Power function operation) Nonlinear Co-excitation Index Taking a value greater than 1 signifies applying a superlinear amplification effect to the product term of joint degradation. This means that when both types of devices simultaneously transition from mild to moderate degradation, the growth rate of off-diagonal elements is much faster than the linear growth of the product term itself. This mathematically characterizes the exponentially amplified collaborative collapse of damping fatigue and battery degradation under dual sub-health conditions; cross-coupling gain constant. It provides a global coupling strength calibration benchmark, which engineers can calibrate based on the actual power overflow transfer relationship between the control valve model of the thermal power unit and the lithium battery system configuration in a specific power plant, so that the off-diagonal response strength of the coupling penalty tensor matches the actual interaction influence between field equipment.

[0069] Furthermore, this tensor structure, which expands adaptively with operating conditions, enables the algorithm to perceive the implicit industrial risk of mutual drag and cascading deterioration between equipment in real time. When subsequently using the coupling penalty tensor to replace the static diagonal weighted matrix in the first embodiment for collaborative criticality index calculation, the off-diagonal elements in the full-rank tensor will introduce cross-product contributions in the quadratic inner product operation, ensuring that the fatigue index is not only calculated through its own... The weighted contribution distance also contributes through the cross-product of the off-diagonal covariance term and the attenuation index, so that the synergistic critical index calculated under the dual deterioration scenario is significantly higher than the square root of the simple sum of squares of the two single-dimensional risks. This provides the control system with a more advanced risk warning signal and effectively suppresses the synergistic deterioration risk caused by power overflow transfer between the mechanical wear of the valve and the degradation of battery life.

[0070] More specifically, in step S3.2.2, the synergistic criticality index is calculated using the coupling penalty tensor on the two-dimensional health status coordinates to obtain the synergistic criticality index. It should be understood that while the coupling penalty tensor can characterize the dynamically increasing coupling relationship between mechanical fatigue and battery electrochemical degradation as the operating conditions change, the coupling penalty tensor itself is still a two-dimensional matrix-type metric object, and the two-dimensional health status coordinates themselves are still a two-dimensional vector-type state object containing fatigue and degradation indices. Subsequent control decision logic needs to determine whether the system has entered a risk range requiring protective adjustment based on a clear numerical threshold. If the aforementioned two-dimensional vector state and matrix-type coupling relationship are not further reduced to a single scalar, it is difficult to directly trigger the generation of subsequent adjustment coefficient vectors and the execution of parameter protection strategies. Therefore, in the technical solution of this application, the synergistic criticality index is further calculated using the coupling penalty tensor on the two-dimensional health status coordinates to obtain the synergistic criticality index. This unifies and compresses the mechanical fatigue risk, battery degradation risk, and their cross-coupling penalties contained in the two-dimensional health status coordinates into a single scalar that can reflect the overall risk level of the system. In this way, the subsequent control decision logic can perform threshold judgment based on the clear value of the collaborative crisis index. Moreover, in the scenario where damper fatigue and battery degradation deteriorate simultaneously, compared with the weighted Euclidean distance based on the static diagonal matrix in the first embodiment, a higher and more sensitive risk measurement result is obtained, thereby providing early warning of systemic collaborative collapse risks that cannot be detected by monitoring only a single dimension.

[0071] Specifically, in a concrete example of this application, the coupling penalty tensor and two-dimensional health status coordinates generated in the previous step are first received. The two-dimensional health status coordinates carry the fatigue index and decay index in column vector form, while the coupling penalty tensor carries the tuning fatigue dimension, battery decay dimension, and the off-diagonal cross-coupling penalty term between them in full-rank second-order symmetric tensor form. Next, the quadratic form of the generalized Mahalanobis distance is used to calculate the quadratic inner product result, which simultaneously includes both single-dimensional self-weight contribution and cross-dimensional cross-product contribution. Then, the square root of the obtained quadratic inner product result is taken to complete the dimensionality reduction mapping from the two-dimensional vector space to the one-dimensional scalar space, yielding the cooperative criticality index, calculated as follows:

[0072]

[0073] in, The collaborative crisis index is calculated at the k-th sampling time. The input is a column vector of a two-dimensional health status coordinate matrix. Let be the transpose row vector of the two-dimensional health status coordinates. This is the coupling penalty tensor generated in the previous step. Unlike the weighted Euclidean distance based on a static diagonal matrix in the first embodiment, the quadratic operation here uses the off-diagonal elements of the full-rank tensor to intertwine the two originally orthogonal and independent health dimensions in the inner product operation—the fatigue index not only through its own... The weighted contribution distance also contributes through a cross-product of the off-diagonal covariance term and the decay index. This means that in a double deterioration scenario, the value of the synergistic crisis index will be significantly higher than the square root of the sum of the squares of the two single-dimensional risks, thus enabling early warning of systemic synergistic collapse risks that cannot be detected by single-dimensional monitoring alone.

[0074] In the above formula, the two-dimensional health status coordinate column vector This vector is used to represent the combined state of the valve's mechanical fatigue and the battery's dynamic degradation at the current sampling moment. Its two components correspond to the fatigue index and the degradation index, respectively; the transposed row vector... The quadratic inner product operation transforms the two-dimensional health status coordinates from column vector form to row vector form that can be left-multiplied by the coupled penalty tensor; the coupled penalty tensor As a benchmark for risk space, its diagonal elements represent the basic sensitivity weight contribution of the damping fatigue dimension and the battery degradation dimension, respectively, while its off-diagonal elements represent the cross-coupling penalty contribution between damping fatigue and battery degradation due to power overflow transfer; quadratic inner product This method can intertwine two originally orthogonal and independent health dimensions in the inner product operation, allowing the fatigue index to contribute distance not only through its own basic sensitivity weight coefficient but also through a cross-product contribution with the decay index via the off-diagonal covariance term. The battery decay dimension also participates in the cross-contribution in the same way. The square root operation restores the quadratic energy metric to a risk scalar with the same dimension as distance, enabling the collaborative criticality index to serve as a direct input for subsequent threshold judgment and activation mapping. Furthermore, unlike the weighted Euclidean distance based on a static diagonal matrix in the first embodiment, the quadratic operation in this step intertwines the two originally orthogonal and independent health dimensions in the inner product operation through the off-diagonal elements in the full-rank tensor, allowing the fatigue index to contribute distance not only through its own weight but also through a cross-product contribution with the decay index. The contribution distance also contributes through a cross-product of the off-diagonal covariance term and the attenuation index. This means that in a dual-deterioration scenario, the value of the collaborative crisis index will be significantly higher than the square root of the sum of squares of the two single-dimensional risks, thus enabling early warning of systemic collaborative collapse risks that cannot be detected by single-dimensional monitoring alone. The output of this collaborative crisis index serves as the input to the subsequent regulation coefficient vector generation stage, allowing the control system to perform threshold activation and saturation mapping on the frequency band offset weight and dead zone amplification weight based on a clear scalar index with coupled risk amplification capabilities. This allows for timely improvement of protective regulation intensity when both the regulator and battery simultaneously enter a sub-healthy state.

[0075] More specifically, in step S3.2.3, the coordinated crisis index is shifted and saturated based on a preset crisis tolerance threshold to obtain the adjustment coefficient vector. It should be understood that, in the actual operation of a power plant, there is a natural trade-off between the economic benefits of frequency regulation services and equipment protection. Prematurely triggering protective parameter adjustments will sacrifice unnecessary frequency regulation benefits, while triggering too late may cause irreversible damage to the equipment. The first embodiment adopts... The mapping function starts generating non-zero adjustment weights when the crisis index is zero, causing the control system to continuously apply weak protective interventions even when the equipment is in normal wear and tear, resulting in unnecessary loss of frequency modulation benefits. Although the cooperative crisis index can comprehensively reflect the combined risk level of tuning fatigue and battery degradation and their cross-coupling amplification effect, the index itself is still a continuously changing scalar. Without introducing a clear threshold judgment mechanism to partition the response, the control system cannot distinguish whether the equipment is in normal wear and tear or has entered a risk range requiring protective intervention, and cannot achieve an optimal balance between economy and safety. Therefore, in the technical solution of this application, the cooperative crisis index is further shifted and saturated based on a preset crisis tolerance threshold to obtain an adjustment coefficient vector. This allows the introduction of a generalized logistic activation function with threshold shift to replace the index mapping in the first embodiment. When the equipment is healthy, protective intervention is completely suppressed to maximize the economic benefits of frequency modulation. When the equipment deteriorates beyond the critical point, a steep step response is used to quickly trigger high-intensity parameter protection adjustments. In this way, the Pareto optimal balance between the economic benefits of power plant frequency regulation and the physical life of key equipment can be fundamentally achieved, effectively avoiding irreversible safety accidents such as gate jamming or battery thermal runaway caused by the lag in control weight response, while eliminating the continuous erosion of frequency regulation economic benefits by unnecessary protective interventions during the normal wear and tear period of equipment.

[0076] Specifically, in a specific example of this application, the coordinated crisis index calculated in the previous step is first received, and preset crisis tolerance thresholds for the frequency regulation loop and dead zone regulation loop are read. The frequency regulation loop crisis tolerance threshold defines the level below which the coordinated crisis index remains at zero for the frequency band offset weight, and the dead zone regulation loop crisis tolerance threshold defines the level below which the coordinated crisis index remains at zero for the dead zone amplification weight. These two thresholds can be set independently based on the physical response characteristics of the frequency regulation loop and the dead zone regulation loop, respectively. Next, the coordinated crisis index is shifted and biased using the preset crisis tolerance thresholds. Specifically, the difference between the coordinated crisis index and the frequency regulation loop crisis tolerance threshold, and the difference between the coordinated crisis index and the dead zone regulation loop crisis tolerance threshold, are calculated respectively. When the coordinated crisis index is below the corresponding crisis tolerance threshold, the difference is negative; when the coordinated crisis index is above the corresponding crisis tolerance threshold, the difference is positive. The significance of this shifting and biasing operation is to shift the response starting point of the activation function from zero to the crisis tolerance threshold, so that the control system does not generate any adjustment drive when the coordinated crisis index does not exceed the threshold. The biased results are then input into two generalized logistic activation functions with independent gain slopes for saturation mapping calculation. The bandwidth offset weights and dead-zone amplification weights of the mapped outputs are concatenated into a column vector to obtain the dynamic adjustment coefficient vector, the calculation formula of which is as follows:

[0077]

[0078] in, This is the vector of dynamic adjustment coefficients generated at the k-th sampling time. The frequency band offset weights are used to adjust the filter cutoff frequency. To adjust the width of the power distribution dead zone, the weight is increased. The input is the collaborative crisis index. and These are the crisis tolerance thresholds for the frequency regulation loop and the dead-zone regulation loop, respectively. This represents the gain slope parameter of the activation function of the frequency adjustment loop. The gain slope parameter represents the activation function of the dead-zone adjustment loop. This represents the physical limit saturation upper limit constant for the frequency adjustment weight. This represents the physical limit saturation upper bound constant for the dead-zone adjustment weight. In the above formula, the shift bias term... and By shifting the inflection point of the S-curve of the logistic function from the origin to its corresponding crisis tolerance threshold, when the coordinated crisis index is below the crisis tolerance threshold, the bias result is negative. After being amplified by the exponential function, the denominator becomes much greater than 1, causing the mapped output to approach zero. The control system does not intervene in the filter parameters and dead zone parameters. The thermal power unit and energy storage system fully respond to the frequency regulation command to maximize economic benefits. Once the coordinated crisis index exceeds the threshold and enters the deterioration range, the bias result turns positive, the denominator rapidly approaches 1, and the mapped output quickly climbs along a steep S-curve to near the saturation value. The control system immediately makes a large adjustment to the filter cutoff frequency and power dead zone with near-full amplitude regulation, achieving hard-core blocking protection for the equipment. Gain slope parameter and The steepness of the S-curve transition near the threshold can be controlled separately for the frequency regulation loop and the dead-zone regulation loop. A larger gain slope parameter makes the transition range of the S-curve narrower near the threshold, allowing the control system to trigger protective regulation with near-full amplitude as soon as the cooperative criticality index exceeds the threshold. A smaller gain slope parameter makes the transition range wider, allowing the control system to gradually increase the regulation intensity near the threshold. Two independent gain slope parameters allow the frequency regulation loop and the dead-zone regulation loop to be set with different trigger sensitivities according to their respective physical response characteristics, avoiding a one-size-fits-all control strategy. Physical limit saturation upper limit constant. and The maximum output amplitude of the frequency band offset weight and the dead zone amplification weight are limited respectively to prevent the adjustment weight from increasing indefinitely under extreme and poor operating conditions, which would cause the filter cutoff frequency or dead zone width to exceed the safe operating range allowed by the system hardware. This ensures that even under the worst dual sub-health conditions, the adjustment range of the control parameters is still constrained within the physically feasible domain.

[0079] Furthermore, the S-curve characteristic of the logistic function endows the control system with a response mode that combines flexible latency and rigid triggering. When the coordinated crisis index is below the crisis tolerance threshold, the function output approaches zero, and the control system does not intervene in the filter parameters and dead zone parameters. The thermal power unit and energy storage system fully respond to the frequency modulation command to maximize economic benefits. Once the coordinated crisis index exceeds the threshold and enters the deterioration range, the function output rapidly climbs along a steep S-curve to near the saturation value. The control system immediately makes a large adjustment to the filter cutoff frequency and power dead zone with near-full amplitude adjustment, realizing hard-core blocking protection for the equipment. After combining the calculated frequency band offset weight and dead zone amplification weight into a dynamic adjustment coefficient vector, it is output to the subsequent adaptive frequency conversion and dead zone parameter update stage. This allows the subsequent stage to adjust the frequency conversion boundary sliding of the high-pass and low-pass filter base frequency and the expansion and contraction of the base dead zone bandwidth based on the weight values ​​in the adjustment coefficient vector. This enables timely improvement of protective adjustment intensity when the regulator and battery simultaneously enter a sub-healthy state, and maximizes the economic return of frequency modulation service when the equipment is healthy.

[0080] Through the second embodiment described above, the algorithm overcomes the blind spot in modeling the coupling relationship between devices using the original static diagonal matrix by introducing a dynamic cross-coupling penalty tensor. This allows the system to perceive in real time the risk of synergistic deterioration caused by power overflow transfer between the mechanical fatigue of the control valve and the electrochemical degradation of the battery. The synergistic criticality index calculated under dual sub-health conditions is significantly higher than the weighted Euclidean distance of the original scheme, thus providing a more advanced and sensitive risk warning signal for control decisions. Simultaneously, the threshold-activated logistic mapping function replaces the original monotonic exponential mapping, eliminating the continuous erosion of frequency regulation economic benefits by unnecessary protective interventions during the normal wear and tear period of the equipment. It maximizes the economic return of frequency regulation services when the equipment is healthy, and rapidly triggers high-intensity parameter protection adjustments with a steep step response when the equipment deteriorates beyond the critical point. This fundamentally achieves a Pareto optimal balance between the economic benefits of power plant frequency regulation and the physical lifespan of key equipment, effectively avoiding irreversible safety accidents such as valve jamming or battery thermal runaway caused by lag in control weight response.

[0081] Specifically, in step S4, based on the frequency band offset weight and dead zone amplification weight in the adjustment coefficient vector, adaptive frequency conversion and dead zone parameter updates are performed on the fundamental frequency and power allocation dead zone bandwidth of the high-pass and low-pass filters combined with the state of charge set to obtain the cutoff frequency parameter and dead zone width parameter. It is understandable that, although the frequency band offset weight and dead zone amplification weight in the adjustment coefficient vector can reflect the adjustment intensity required by the current joint health status of the equipment, the vector itself is only a set of abstract control weight scalars and has not yet been transformed into physical filter parameters and power dead zone parameters that can directly act on the hierarchical power decoupling link. The subsequent hierarchical frequency domain decoupling and command truncation filtering links need to receive cutoff frequency values ​​and dead zone bandwidth values ​​with clear physical dimensions as input. At the same time, the adjustment of the filter cutoff frequency and dead zone bandwidth depends not only on the adjustment weight driven by the equipment health status, but also on the current actual available capacity of the energy storage system. If the state of charge of a certain energy storage unit is close to the charge and discharge boundary, even if the health status weight requires more high-frequency commands to be transferred to the energy storage unit, it cannot physically continue to undertake additional power throughput tasks. If the remaining capacity margin of each energy storage unit in the state of charge is not included in the parameter update calculation, it may lead to a contradiction between the sliding direction of the cutoff frequency and the actual carrying capacity of the energy storage system. Therefore, in the technical solution of this application, based on the frequency band offset weight and dead zone amplification weight in the adjustment coefficient vector, adaptive frequency conversion and dead zone parameter updates are performed on the fundamental frequency and power allocation dead zone bandwidth of the high-pass and low-pass filters combined with the state of charge concentration to obtain cutoff frequency parameters and dead zone width parameters. This integrates the abstract health status adjustment weight with the actual capacity constraints of the energy storage system, generating control parameters with clear physical dimensions that simultaneously meet equipment protection requirements and capacity feasibility constraints. In this way, the filter cutoff frequency and power dead zone bandwidth used in the subsequent layered frequency domain decoupling stage can be updated synchronously with the changes in equipment health status and energy storage capacity status in each sampling period. When the equipment is healthy and energy storage is sufficient, the power allocation boundary close to the fundamental parameters is maintained to maximize the frequency modulation response capability. When the equipment deteriorates or energy storage approaches the boundary, the frequency division bandwidth and dead zone range are adaptively adjusted to implement protective power redistribution.

[0082] More specifically, in this embodiment, step S4 includes: S4.1: Extracting the frequency band offset weight in the adjustment coefficient vector, and combining it with the remaining capacity margin of each energy storage unit in the state of charge set that deviates from the ideal center value as a multiplication operator, performing frequency conversion boundary sliding based on health status weight on the preset high-pass and low-pass filter base frequency to obtain the cutoff frequency parameter; S4.2: Extracting the dead zone amplification weight in the adjustment coefficient vector, and combining it with the capacity extreme value of the energy storage unit with the largest deviation in the state of charge set, scaling the preset base dead zone bandwidth to obtain the dead zone width parameter.

[0083] Specifically, in a specific example of this application, the adjustment coefficient vector is first unpacked and extracted to obtain the frequency band offset weight as the adjustment factor for driving the cutoff frequency sliding. At the same time, the state of charge set is analyzed to extract the real-time state of charge values ​​of the lithium battery and the supercapacitor, and the remaining capacity margin of each energy storage unit deviating from the ideal center value is calculated. The remaining capacity margin is calculated by taking the absolute value of the deviation between the state of charge of each energy storage unit and the ideal center value and subtracting it from 1, so that the margin is maximized when the state of charge of the energy storage unit is at the ideal center value and approaches zero when the state of charge approaches the charge-discharge boundary. Next, using the system's preset low-pass filter and high-pass filter base cutoff frequencies as benchmarks, the frequency band offset weight is multiplied by the remaining capacity margin of the corresponding energy storage unit as a multiplication operator. This multiplication operator slides the low-pass cutoff frequency downwards and the high-pass cutoff frequency upwards. When the frequency band offset weight increases, indicating a deterioration in equipment health, the low-pass cutoff frequency decreases, causing more intermediate frequency components to shift from the thermal power unit side to the energy storage side. The high-pass cutoff frequency increases, causing the supercapacitor to only handle extremely high frequency pulses within a narrower frequency band. When the remaining capacity margin of the corresponding energy storage unit decreases, the multiplication operator automatically reduces the sliding amplitude to prevent power commands exceeding the energy storage capacity from being forcibly allocated to energy storage units with insufficient capacity. The calculated low-pass and high-pass cutoff frequencies are packaged to generate cutoff frequency parameters, the calculation formula of which is:

[0084]

[0085] in, This represents the cutoff frequency parameter calculated at the k-th sampling time. This indicates the updated cutoff frequency value of the low-pass filter. This indicates the updated cutoff frequency value of the high-pass filter. This represents the preset fundamental cutoff frequency constant of the low-pass filter under ideal healthy conditions. This represents the fundamental cutoff frequency constant of the high-pass filter preset under ideal healthy conditions. This represents the frequency band offset weight extracted from the adjustment coefficient vector. This represents the real-time state of charge (SOC) value of the lithium battery extracted from the SOC set. This represents the real-time state of charge (SOC) value of the supercapacitor extracted from the SOC set. In the above formula, the update expression for the low-pass cutoff frequency... Multiplication operators in It can be simultaneously constrained by both the health status adjustment weight and the remaining capacity margin of the lithium battery, among which This determines the proportion of intermediate frequency commands that need to be transferred from the thermal power unit side to the energy storage side due to equipment degradation. This determines whether the lithium battery currently has the capacity to receive these transfer commands. When the lithium battery's state of charge is at the ideal center value of 0.5, the margin term takes the maximum value of 1, so that the frequency slippage is entirely determined by the health status weight. When the lithium battery's state of charge approaches the boundary of full charge or full discharge, the margin term approaches zero to automatically suppress frequency slippage to avoid imposing excessive power burden on lithium batteries with insufficient capacity. In the update expression of the high-pass cutoff frequency, the state of charge margin of the supercapacitor is used as a constraint operator. Its significance is that when the supercapacitor capacity is sufficient, the high-pass cutoff frequency is allowed to move up sufficiently to narrow the response frequency band of the supercapacitor and thus reduce its power burden. When the supercapacitor capacity approaches the boundary, the upward movement is automatically limited to preserve its basic response capability to ultra-high frequency pulses. Then, the adjustment coefficient vector is unpacked and extracted a second time to obtain the dead zone amplification weight. Simultaneously, the state of charge (SCC) set is called again, and the item with the largest deviation between the SCC deviation of the lithium battery and the supercapacitor is selected as the capacity extremum constraint factor. Using the system's preset basic dead zone bandwidth as a benchmark, the dead zone amplification weight and the capacity extremum constraint factor are used to scale the basic dead zone bandwidth through a nonlinear exponential expansion function. When the dead zone amplification weight increases and the energy storage capacity approaches the boundary, the dead zone bandwidth is significantly widened to filter out more high-frequency, small-power glitches that could cause equipment malfunctions. The dead zone width parameter is generated, and its calculation formula is as follows:

[0086]

[0087] in, This represents the dead zone width parameter calculated at the k-th sampling time. This represents the preset base dead-zone bandwidth constant when the system is in good health and has sufficient energy storage. This represents the dead zone amplification weight extracted from the adjustment coefficient vector. The adjustment sensitivity constant represents the nonlinear expansion of the dead zone. This represents the function that takes the maximum value. This indicates the degree to which the state of charge of a lithium battery deviates from the ideal center value. This indicates the degree to which the state of charge of the supercapacitor deviates from the ideal center value. In the above formula, the maximum value function selects the one with the worst capacity state from the two types of energy storage units as the dominant constraint factor for dead zone expansion. This ensures that the adjustment of the dead zone width is based on the capacity state of the weakest link, avoiding the situation where a narrow dead zone is maintained when an energy storage unit is already close to its capacity boundary, thus preventing invalid commands from continuously impacting that unit; the exponential expansion function... The rate of increase in dead zone width accelerates with the degree of deviation from the energy storage capacity. When the energy storage capacity is near the ideal center value, the exponential term is close to 1, so the dead zone width is only linearly adjusted by the dead zone amplification weight. When the energy storage capacity deviates significantly from the center value, the exponential term increases sharply, causing the dead zone width to expand rapidly to significantly filter high-frequency micro-instructions. (Sensitivity constant) The steepness of this nonlinear expansion process was controlled. Furthermore, the calculated cutoff frequency parameters and dead zone width parameters were synchronously output to the subsequent hierarchical power decoupling and command issuance stages. This enabled the subsequent stages to perform dead zone truncation filtering and hierarchical frequency domain decoupling on effective frequency modulation commands based on these two sets of dynamic parameters constrained by both health status weights and energy storage capacity margins. This achieved real-time coordinated migration of the power allocation boundary according to the equipment health status and energy storage capacity status.

[0088] Specifically, in step S5, the preprocessed frequency modulation command is filtered for high-frequency glitches using the dead zone width parameter to obtain an effective frequency modulation command, and the effective frequency modulation command is decoupled into a layered frequency domain based on the cutoff frequency parameter to obtain the thermal power command, battery power command, and supercapacitor power command. It is understandable that although the cutoff frequency parameter and dead zone width parameter have been adaptively updated in the previous steps based on the equipment health status weight and energy storage capacity margin, these two sets of parameters are only intermediate variables at the control strategy level and have not yet been actually applied to the physical allocation process of the frequency regulation command. The ultimate goal of the joint frequency regulation system is to break down the frequency regulation command issued by the grid into power sub-commands that thermal power units, lithium batteries, and supercapacitors can safely execute. If the above dynamic parameters are not injected into the actual command truncation and frequency domain decoupling calculation, the protective regulation intention generated in the previous steps based on health status assessment and capacity constraints will not be implemented as actual power allocation control of physical equipment. At the same time, if the high-frequency small power spikes that still remain in the preprocessed frequency regulation command and fall within the dead zone are not truncated, they will be allocated to thermal power units or energy storage systems in the subsequent frequency domain decoupling, resulting in ineffective oscillation of the regulator and ineffective micro-circulation of the battery. Therefore, in the technical solution of this application, the dead-zone width parameter is further used to perform high-frequency glitch truncation filtering on the preprocessed frequency modulation command to obtain an effective frequency modulation command. Furthermore, based on the cutoff frequency parameter, the effective frequency modulation command is decoupled into layers in the frequency domain to obtain the thermal power command, battery power command, and supercapacitor power command. This transforms the dynamic control parameters generated in the preceding steps into actual physical allocation actions for the frequency modulation command, completing a full closed loop from health status perception to power allocation execution. In this way, thermal power units can receive only low-frequency reference power commands that match their thermodynamic ramp-up characteristics, lithium batteries can only undertake the task of medium-frequency energy compensation after dead-zone filtering and frequency band limitation, and supercapacitors can only respond to extremely high-frequency transient pulses. Thus, while ensuring the grid frequency regulation assessment indicators, the protective allocation strategy determined in the preceding steps based on equipment health status and energy storage capacity constraints is effectively implemented in the power command output of each sampling cycle.

[0089] Figure 5 This is a flowchart of step S5 of the frequency regulation method for hybrid energy storage thermal power units based on hierarchical adaptive control according to an embodiment of this application. Figure 5As shown, step S5 includes: S5.1: Based on the dead-zone width parameter, the preprocessed frequency modulation command is truncated and effective features are extracted under controlled dead-zone conditions to obtain an effective frequency modulation command; S5.2: Based on the low-pass cutoff frequency component in the cutoff frequency parameter, the effective frequency modulation command is subjected to low-pass frequency conversion filtering and decoupling from the thermal power reference power to obtain the thermal power power command and the medium-to-high frequency residual power command; S5.3: Based on the high-pass cutoff frequency component in the cutoff frequency parameter, the medium-to-high frequency residual power command is subjected to ultra-high frequency pulse stripping to obtain the supercapacitor power command, and the supercapacitor power command is subtracted from the medium-to-high frequency residual power command to obtain the lithium battery power command.

[0090] Accordingly, in step S5.1, based on the dead-time width parameter, the preprocessed frequency modulation command is truncated and effective features are extracted under controlled dead-time conditions to obtain an effective frequency modulation command. It is understandable that although the preprocessed frequency modulation command has been smoothed and filtered to remove white noise interference in the previous steps, it still contains a large number of high-frequency small power fluctuation components with extremely small amplitudes and frequent changes. The amplitude of these small fluctuation components is insufficient to drive the thermal power unit or energy storage system to generate effective power output, but they will be allocated to the actuators of each physical device in the subsequent layered frequency domain decoupling process. This causes the turbine control valve to perform ineffective high-frequency reciprocating oscillations within a very small opening range and the lithium battery to perform ineffective micro-cycle switching at a very shallow charge and discharge depth. The dead zone width parameter has been adaptively updated in the previous steps based on the equipment health status weight and the extreme value constraint of the energy storage capacity. Its value directly reflects the system's tolerance boundary for high-frequency small commands under the current operating conditions. If the preprocessed frequency modulation command is not truncated and filtered using this dynamic dead zone parameter, the protective anti-operation intention determined in the previous steps based on the health status assessment will not be executed at the command level. Therefore, in the technical solution of this application, the preprocessed frequency modulation command is further truncated and its effective features are extracted under controlled dead-zone conditions based on the dead-zone width parameter to obtain an effective frequency modulation command. This eliminates high-frequency micro-power glitches that would cause ineffective equipment operation at the source before layered frequency domain decoupling, retaining only the command components that can drive the physical equipment to generate effective work. In this way, the effective frequency modulation command received in the subsequent layered frequency domain decoupling stage will no longer contain ineffective disturbance components falling within the dynamic dead-zone range, thereby avoiding ineffective mechanical wear of the tuning gate due to responding to micro-commands and ineffective electrochemical throughput of the lithium battery due to responding to micro-commands. At the same time, since the dead-zone width parameter changes dynamically with the health status of the equipment and the energy storage capacity, the truncation filtering strength can maintain a narrow dead zone when the equipment is healthy to maximize the frequency modulation response capability, and adaptively widen the dead zone when the equipment deteriorates to enhance the protective rejection force.

[0091] Specifically, in a specific example of this application, the instantaneous power amplitude of the preprocessed frequency modulation command and the current value of the dead-zone width parameter are first synchronously read at each discrete sampling time. The preprocessed frequency modulation command originates from the output of the original frequency modulation command after smoothing and filtering in the preceding feature sequence extraction step, and the dead-zone width parameter originates from the output calculated based on the dead-zone amplification weight and energy storage capacity extreme value constraint in the preceding adaptive frequency conversion and dead-zone parameter update steps. Next, the absolute value of the instantaneous power amplitude of the preprocessed frequency modulation command is taken and compared with the dead-zone width parameter to determine whether the command amplitude at the current time falls within the cutoff interval defined by the dead-zone width parameter. If the absolute amplitude of the preprocessed frequency modulation command is less than or equal to the dead-zone width parameter, the command is determined to be a high-frequency micro-power glitch that cannot drive the physical equipment to generate effective work. The output value of the effective frequency modulation command is forcibly set to zero, so that neither the thermal power unit nor the energy storage system responds to the command during the sampling period. Then, for cases where the absolute amplitude of the preprocessed frequency modulation command is greater than the dead-time width parameter, the product of the dead-time width parameter and the command direction sign is subtracted from the original amplitude of the preprocessed frequency modulation command to eliminate the influence of the dead-time offset on the effective command amplitude. This ensures that the effective frequency modulation command changes continuously from zero, avoiding power step jumps at the dead-time boundary. The calculation formula is as follows:

[0092]

[0093] in, This represents the instantaneous power amplitude of the effective frequency modulation command calculated at the k-th sampling time. This represents the instantaneous power amplitude of the preprocessed frequency modulation command input at the k-th sampling time. This represents the dead zone width parameter input at the k-th sampling time. This represents the absolute value operator. This represents a sign function that returns positive one when the input variable is greater than zero and negative one when the input variable is less than zero. In the formula above, the conditional judgment... The significance of this scenario lies in using the dead zone width parameter as a dynamic threshold boundary to identify minute fluctuation components in the pre-processed frequency modulation command whose amplitude is too small to drive physical equipment to generate effective power output. These minute fluctuation components in the joint frequency modulation scenario correspond to random perturbations of the grid frequency near steady state. Their amplitude is insufficient to overcome the static friction of the regulating valve core and the dead zone gap of the hydraulic actuator. If transmitted to the actuator, they will only cause ineffective chattering of the regulating valve within the dead zone gap. Forcing it to zero can eliminate this type of ineffective mechanical impact from the source; deduction item Capable of bias compensation for valid instructions that cross the dead zone boundary, sign function To ensure the deduction direction aligns with the charging and discharging direction of the pre-processed frequency modulation command, the effective frequency modulation command starts from zero at the dead zone boundary rather than abruptly changing from the dead zone width value. This continuous processing avoids secondary power oscillations in thermal power units and energy storage systems caused by sudden changes in power commands near the dead zone boundary. Furthermore, since the dead zone width parameter is dynamically updated in each sampling period based on the equipment health status and energy storage capacity, when the equipment is in a healthy state and the energy storage capacity is sufficient, the dead zone width parameter remains near a small base value, truncating only extremely small noise level commands to maximize the preservation of frequency modulation response capability. When the equipment health status deteriorates or the energy storage capacity approaches the boundary, the dead zone width parameter adaptively increases, and the truncation range is correspondingly widened. More high-frequency commands with small to medium amplitudes are rejected to reduce the impact burden on deteriorated equipment. The calculated effective frequency modulation command is output to the subsequent low-pass frequency conversion filter and thermal power reference power decoupling stage, ensuring that the command signal processed by the subsequent layered frequency domain decoupling no longer contains tiny power spikes that would cause the equipment to malfunction.

[0094] Accordingly, in step S5.2, based on the low-pass cutoff frequency component in the cutoff frequency parameter, the effective frequency modulation command is subjected to low-pass frequency conversion filtering and decoupled from the thermal power reference power to obtain the thermal power command and the medium- and high-frequency residual power command. It is understandable that although the effective frequency regulation command has eliminated high-frequency small power glitches through dead-zone truncation filtering in the previous steps, it still contains wide-band power fluctuation components from low frequency to high frequency. However, thermal power units are constrained by the inherent large inertia and time constant of the boiler-turbine thermal system and can only track slowly changing low-frequency power commands. If the effective frequency regulation command containing mid-to-high frequency fluctuation components is directly issued to the thermal power unit, the turbine control valve will be forced to respond to rapid fluctuations that exceed its thermodynamic ramp-up capability. This will not only fail to generate effective power output but will also exacerbate the mechanical fatigue accumulation of the control valve. At the same time, the low-pass cutoff frequency component in the cutoff frequency parameter has been adaptively updated in the previous steps based on the equipment health status weight and the remaining capacity margin of the lithium battery. Its value directly determines the power sharing frequency boundary between the thermal power unit and the energy storage system. If this dynamic low-pass cutoff frequency component is not injected into the actual filtering calculation, the frequency band offset intention determined in the previous steps based on the health status assessment will not be executed at the power allocation level. Therefore, in the technical solution of this application, the effective frequency modulation command is further decoupled from the thermal power reference power by low-pass frequency conversion filtering based on the low-pass cutoff frequency component in the cutoff frequency parameter to obtain the thermal power command and the medium-to-high frequency residual power command. This allows the low-frequency reference power flow that matches the thermodynamic response characteristics of the thermal power unit in the effective frequency modulation command to be extracted as the thermal power command, and the medium-to-high frequency fluctuation component that the thermal power unit cannot track to be separated as the medium-to-high frequency residual power command for subsequent energy storage systems. In this way, the thermal power unit can receive only the smooth low-frequency command after dynamic low-pass filtering, and its rate of change is constrained within a range that matches the ramp-up capability of the boiler-turbine thermal system. This avoids ineffective mechanical wear caused by rapid fluctuations in response that exceed the physical capacity of the control valve. At the same time, since the low-pass cutoff frequency component changes dynamically with the health status of the equipment, when the control valve fatigue intensifies, the low-pass cutoff frequency adaptively decreases, making the thermal power command smoother and the control valve action amplitude further reduced. When the equipment is healthy, the low-pass cutoff frequency is maintained near a higher base value, allowing the thermal power unit to undertake a wider frequency band of frequency modulation tasks to maximize the frequency modulation response capability.

[0095] Specifically, in a specific example of this application, the low-pass cutoff frequency component is first extracted from the cutoff frequency parameter and converted into the corresponding discrete low-pass filter time constant. This conversion is based on the standard physical relationship between the cutoff frequency and time constant of a first-order low-pass filter. The smaller the low-pass cutoff frequency component, the larger the corresponding time constant, and the stronger the smoothing effect of the filter on the input signal. The conversion formula is as follows:

[0096]

[0097] in, This represents the discrete low-pass filter time constant obtained by converting the low-pass cutoff frequency component at the k-th sampling time. This represents the low-pass cutoff frequency component extracted from the cutoff frequency parameter. Next, the effective frequency modulation command is input to a first-order discrete low-pass filter with the aforementioned dynamic time constant for frequency domain decoupling. At each discrete sampling time, this filter performs a weighted linear combination of the instantaneous power amplitude of the current effective frequency modulation command and the thermal power command from the previous sampling time. The weighting coefficient is determined by the ratio of the discrete sampling time step to the dynamic time constant, extracting the slowly changing low-frequency reference power flow as the thermal power command. Its calculation formula is:

[0098]

[0099] in, This represents the thermal power command calculated and generated at the k-th sampling time. This represents the discrete sampling time step of the system. This represents the discrete low-pass filter time constant at the k-th sampling time. This represents the instantaneous power amplitude of the valid frequency modulation command input at the k-th sampling time. This represents the thermal power command at the previous sampling time. In the above formula, the weighting coefficient... This coefficient is used to control the immediate contribution ratio of the current effective frequency modulation command to the thermal power command output. Its value decreases as the time constant increases, meaning that when equipment health deteriorates, leading to a decrease in the low-pass cutoff frequency and an increase in the time constant, the impact of rapid fluctuations in the current effective frequency modulation command on the thermal power command is further suppressed, and the command received by the thermal power unit changes more slowly; weighting coefficient The coefficient used to control the inertia retention ratio of the thermal power command at the previous sampling moment to the current output increases with the increase of the time constant, meaning that the thermal power command relies more on historical output values ​​for smooth continuation, further reducing the amplitude and frequency of the regulating valve's operation, thereby alleviating the mechanical fatigue burden on the regulating valve under high-frequency modulation conditions. The sum of the two weighting coefficients is always equal to one, ensuring the power conservation characteristic of the filtering operation, that is, all power information in the effective frequency modulation command will not generate gain or attenuation during the filtering process, but only undergo frequency domain redistribution. Then, using the power conservation principle, the effective frequency modulation command is subtracted from the thermal power command already allocated to the thermal power unit to obtain the medium- and high-frequency fluctuation components that the thermal power unit cannot track due to the limitations of its thermodynamic response capability, generating the medium- and high-frequency residual power command, the calculation formula of which is:

[0100]

[0101] in, This represents the mid-to-high frequency residual power command calculated at the k-th sampling time. This represents the valid frequency modulation command input at the k-th sampling time. This represents the thermal power command allocated at the k-th sampling time. In the above formula, the subtraction operation... This ensures that all valid frequency modulation commands are allocated without omission. The low-frequency components extracted by the low-pass filter are handled by the thermal power unit, while the remaining mid-to-high frequency fluctuation components are all incorporated into the mid-to-high frequency residual power command, awaiting secondary decoupling between the lithium battery and the supercapacitor. This power conservation subtraction method guarantees that at any sampling moment, the sum of the thermal power command and the mid-to-high frequency residual power command is strictly equal to the valid frequency modulation command, preventing power deficits or overflows due to the frequency domain allocation process. Furthermore, the calculated thermal power command is sent to the underlying actuator of the thermal power unit, and the mid-to-high frequency residual power command is output to the subsequent ultra-high frequency pulse stripping stage. This enables the subsequent stage to perform secondary power decoupling between the lithium battery and the supercapacitor based on the high-pass cutoff frequency component in the cutoff frequency parameter.

[0102] Accordingly, in step S5.3, based on the high-pass cutoff frequency component in the cutoff frequency parameter, the mid-to-high frequency residual power command is stripped of ultra-high frequency pulses to obtain the supercapacitor power command, and the supercapacitor power command is subtracted from the mid-to-high frequency residual power command to obtain the lithium battery power command. It should be understood that the mid-to-high frequency residual power command is all the mid-to-high frequency fluctuation components remaining after deducting the low-frequency reference power flow undertaken by the thermal power unit from the effective frequency regulation command. This includes both mid-frequency power fluctuations with moderate change rates and long durations, as well as ultra-high frequency power pulse spikes with extremely fast change rates and extremely short durations caused by transient events such as sudden load disturbances or short-circuit fault recovery on the grid side. Lithium batteries and supercapacitors have fundamental differences in physical characteristics. Lithium batteries have high energy density and moderate power density, making them suitable for undertaking mid-frequency energy compensation tasks with long durations, while supercapacitors have extremely high power density and millisecond-level response speeds. However, the energy density is limited, making it suitable for the rapid absorption and release of transient ultra-high frequency pulses. If the mid-to-high frequency residual power command is indiscriminately allocated to either the lithium battery or the supercapacitor, the lithium battery will be forced to respond to ultra-high frequency pulses exceeding its power density capacity, accelerating electrochemical degradation, or the supercapacitor will be forced to undertake continuous mid-frequency tasks exceeding its energy capacity, rapidly depleting its state of charge. Meanwhile, the high-pass cutoff frequency component in the cutoff frequency parameter has already been adaptively updated in the previous steps based on the device health status weight and the remaining capacity margin of the supercapacitor; its magnitude directly determines the power sharing frequency boundary between the lithium battery and the supercapacitor. Therefore, in the technical solution of this application, the mid-to-high frequency residual power command is further stripped of ultra-high frequency pulses based on the high-pass cutoff frequency component in the cutoff frequency parameter to obtain the supercapacitor power command, and the supercapacitor power command is subtracted from the mid-to-high frequency residual power command to obtain the lithium battery power command. This decouples the mid-to-high frequency residual power command between the lithium battery and the supercapacitor according to its frequency domain characteristics, allowing each energy storage unit to undertake only the frequency band tasks matching its physical characteristics. In this way, the supercapacitor can respond only to the extremely high frequency transient pulses after dynamic high-pass filtering, while the lithium battery can only bear the smooth intermediate frequency fluctuation components remaining after the extremely high frequency pulses are stripped away. When the health of the equipment deteriorates, the high-pass cutoff frequency is adaptively increased to further narrow the response frequency band of the supercapacitor to reduce its power burden. When the equipment is healthy, the high-pass cutoff frequency is maintained near a low base value, allowing the supercapacitor to undertake high-frequency tasks in a wider frequency band to give full play to its fast response advantage.

[0103] Specifically, in a specific example of this application, the high-pass cutoff frequency component is first extracted from the cutoff frequency parameter and converted into the corresponding discrete high-pass filter time constant. This conversion is based on the standard physical relationship between the cutoff frequency and time constant of a first-order high-pass filter. The higher the high-pass cutoff frequency component, the smaller the corresponding time constant. The filter only allows extremely high-frequency components with faster rates of change to pass through. The conversion formula is as follows:

[0104]

[0105] in, This represents the discrete high-pass filter time constant obtained by converting the high-pass cutoff frequency component at the k-th sampling time. This represents the high-pass cutoff frequency component extracted from the cutoff frequency parameter. Next, the mid-to-high frequency residual power command is input to a first-order discrete high-pass filter with the aforementioned dynamic time constant for ultra-high frequency pulse stripping. This filter performs recursive calculations at each discrete sampling time based on the differential change in the mid-to-high frequency residual power command between the current and previous sampling times, as well as the supercapacitor power command from the previous sampling time. It extracts only the ultra-high frequency transient pulse component with the fastest changing rate from the mid-to-high frequency residual power command as the supercapacitor power command. Its calculation formula is:

[0106]

[0107] in, This represents the supercapacitor power command calculated at the k-th sampling time. This represents the discrete high-pass filter time constant at the k-th sampling time. This represents the discrete sampling time step of the system. This indicates the supercapacitor power command at the previous sampling time. This represents the mid-to-high frequency residual power command input at sampling time kk. This represents the mid-to-high frequency residual power command at the previous sampling time. In the above formula, the difference term... This is used to extract the instantaneous change in the mid-to-high frequency residual power command between adjacent sampling periods. This change directly reflects the drastic change in the high-frequency components of the mid-to-high frequency residual power command. The ultra-high frequency pulse spikes caused by sudden load disturbances on the grid side are represented by large-amplitude positive and negative jumps in this differential term, while the relatively slowly changing mid-frequency fluctuation components are represented by small-amplitude gradual changes in this differential term, thus achieving preliminary separation of ultra-high frequency components and mid-frequency components at the differential level; weighting coefficients This coefficient, used to control the proportion of differential signals passed through the high-pass filter, decreases as the time constant decreases. This means that when equipment health deteriorates, leading to an increase in the high-pass cutoff frequency and a decrease in the time constant, the filter's proportion of differential signals passed through decreases, allowing only extreme pulses with faster change rates to pass through. The supercapacitor's response frequency band is further narrowed to reduce its power burden. (Recursive term) This maintains the time continuity of the high-pass filter output, allowing the supercapacitor power command to transition smoothly between adjacent sampling periods without abrupt jumps. Then, using the power balance equation, the supercapacitor power command already allocated to the supercapacitor is subtracted from the residual mid-to-high frequency power command, yielding the smoothed mid-frequency fluctuation component remaining after stripping away the ultra-high frequency pulse spikes. This smoothed mid-frequency fluctuation component is then used to generate the lithium battery power command, calculated using the following formula:

[0108]

[0109] in, This represents the lithium battery power command calculated at the k-th sampling time. This represents the mid-to-high frequency residual power command input at the k-th sampling time. This represents the supercapacitor power command allocated at the k-th sampling time. In the above formula, the subtraction operation... This ensures that the mid-to-high frequency residual power command is distributed between the lithium battery and the supercapacitor without any omission. The ultra-high frequency pulse component extracted by the high-pass filter is handled by the supercapacitor, while the remaining mid-frequency fluctuation component is handled by the lithium battery. The mid-frequency component has a moderate rate of change and a long duration, which matches the electrochemical characteristics of the lithium battery, which has both a certain power density and a high energy density. This power conservation subtraction method ensures that at any sampling moment, the sum of the supercapacitor power command and the lithium battery power command is strictly equal to the mid-to-high frequency residual power command. Combined with the conservation relationship in the previous step that the sum of the thermal power command and the mid-to-high frequency residual power command is strictly equal to the effective frequency modulation command, the sum of the three power commands is strictly equal to the effective frequency modulation command across the entire frequency band, and there will be no power deficit or power overflow due to the multi-level frequency domain allocation process. Furthermore, the calculated power commands for the supercapacitor and lithium battery are sent to the corresponding underlying power converter actuators. This completes the entire closed-loop control link from the acquisition of the original frequency regulation command, equipment health status assessment, dynamic parameter generation, dead zone truncation filtering to three-level hierarchical frequency domain decoupling. This allows the thermal power unit, lithium battery, and supercapacitor to each undertake only the frequency band tasks that match their physical characteristics. Moreover, the power allocation boundary migrates in real time in each sampling period as the equipment health status and energy storage capacity change. Under the premise of ensuring the grid frequency regulation assessment indicators, global coordinated protection against mechanical wear of the regulator and battery life degradation is achieved.

[0110] In summary, the frequency regulation method for hybrid energy storage thermal power units based on hierarchical adaptive control according to the embodiments of this application is explained. It extracts feature sequences from real-time collected valve travel data, battery operation data, supercapacitor charge data, and frequency regulation commands. Based on this, it performs mechanical fatigue wear evolution on the cumulative travel of the valve and quantifies the micro-cycle equivalent capacity loss of the battery discharge depth to obtain fatigue index and attenuation index. Then, it maps the two to two-dimensional health status coordinates and uses a fuzzy logic inference engine to perform potential field optimization and inference to generate an adjustment coefficient vector. Using this vector in combination with the energy storage charge state, it adaptively updates the filter cutoff frequency and power allocation dead zone. Finally, based on the updated parameters, it performs dead zone truncation filtering and hierarchical frequency domain decoupling on the frequency regulation command, and outputs the power commands of thermal power, battery, and supercapacitor respectively, thereby achieving global collaborative protection of valve wear and battery life while ensuring frequency regulation performance.

[0111] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A frequency regulation method for hybrid energy storage thermal power units based on hierarchical adaptive control, characterized in that, include: S1: Extract feature sequences from the real-time collected valve travel data, battery operation data, supercapacitor charge data, and frequency modulation commands to obtain travel feature sequences, discharge depth sequences, charge state sets, and preprocessed frequency modulation commands; S2: Mechanical fatigue damage evolution and microcirculation equivalent capacity loss quantification are performed on the travel characteristic sequence and discharge depth sequence to obtain fatigue index and attenuation index; S3: Construct a two-dimensional health status coordinate system based on the fatigue index and the decay index, and use a multi-input multi-output fuzzy logic reasoning engine to perform potential field optimization deduction on the two-dimensional health status coordinate system to obtain the adjustment coefficient vector. S4: Based on the frequency band offset weight and dead zone amplification weight in the adjustment coefficient vector, adaptive frequency conversion and dead zone parameter update are performed on the fundamental frequency and power allocation dead zone bandwidth of the high-pass and low-pass filters combined with the state of charge set to obtain the cutoff frequency parameter and dead zone width parameter. S5: High-frequency glitches are truncated and filtered in the preprocessed frequency modulation command using the dead zone width parameter to obtain the effective frequency modulation command. The effective frequency modulation command is then decoupled in the hierarchical frequency domain based on the cutoff frequency parameter to obtain the thermal power command, battery power command and supercapacitor power command.

2. The frequency regulation method for hybrid energy storage thermal power units based on hierarchical adaptive control according to claim 1, characterized in that, Step S1 includes: S1.1: Resample and align the valve travel data, battery operation data, and supercapacitor charge data to obtain aligned valve travel data, aligned battery status data, and aligned supercapacitor charge data; perform smoothing filtering on the frequency modulation command to obtain a preprocessed frequency modulation command. S1.2: Extract stroke feature sequences from the aligned valve stroke data; S1.3: Unpack the aligned battery state data to separate the current sub-item and the state of charge sub-item, and perform discharge depth throughput conversion on the current sub-item to obtain the discharge depth sequence; S1.4: Perform dimensional concatenation and vectorization on the aligned battery charge data and the aligned supercapacitor charge data to obtain the charge state set.

3. The frequency regulation method for hybrid energy storage thermal power units based on hierarchical adaptive control according to claim 2, characterized in that, S1.2, extract the stroke feature sequence from the aligned valve stroke data using the following formula: ； in, For the first The instantaneous percentage of the aligned turbine control valve stroke opening data input at any given time; For the first The instantaneous percentage of the aligned turbine control valve stroke opening data input at any given time. For the first The cumulative travel value in the cumulative travel feature sequence of the output at any time.

4. The frequency regulation method for hybrid energy storage thermal power units based on hierarchical adaptive control according to claim 2, characterized in that, S1.3 includes: calculating the discharge depth throughput of the current sub-term using the following formula: ； in, For the first The instantaneous current value of the lithium battery obtained from constant unpacking. The discrete sampling time step of the system, This refers to the rated nominal capacity of the lithium battery system.

5. The frequency regulation method for hybrid energy storage thermal power units based on hierarchical adaptive control according to claim 1, characterized in that, Step S2 includes: S2.1: Input the travel characteristic sequence into the rainflow counting method fatigue life model to obtain the fatigue index; S2.2: Input the discharge depth sequence into the electrochemical Arrhenius capacity decay model to obtain the decay index.

6. The frequency regulation method for hybrid energy storage thermal power units based on hierarchical adaptive control according to claim 1, characterized in that, Step S3 includes: S3.1: Based on the preset two-dimensional Euclidean space, the fatigue index is used as the first dimension and the decay index is used as the second dimension to perform orthogonal combination and vectorization mapping to obtain two-dimensional health status coordinates. S3.2: Input the two-dimensional health status coordinates into the multi-input multi-output fuzzy logic inference engine to activate the frequency band offset weight and dead zone amplification weight respectively to perform potential field optimization inference to obtain the adjustment coefficient vector.

7. The frequency regulation method for hybrid energy storage thermal power units based on hierarchical adaptive control according to claim 1, characterized in that, Step S4 includes: S4.1: Extract the frequency band offset weight in the adjustment coefficient vector, and combine it with the remaining capacity margin of each energy storage unit in the state of charge set that deviates from the ideal center value as a multiplication operator to perform frequency conversion boundary sliding based on health status weight on the preset high and low pass filter base frequency to obtain the cutoff frequency parameter. S4.2: Extract the dead zone amplification weight from the adjustment coefficient vector, and combine it with the capacity extreme value of the energy storage unit with the largest deviation of the state of charge concentration, to stretch and expand the preset basic dead zone bandwidth to obtain the dead zone width parameter.

8. The frequency regulation method for hybrid energy storage thermal power units based on hierarchical adaptive control according to claim 1, characterized in that, Step S5 includes: S5.1: Based on the dead-time width parameter, the preprocessed frequency modulation command is truncated and effective features are extracted under controlled dead-time conditions to obtain an effective frequency modulation command; S5.2: Based on the low-pass cutoff frequency component in the cutoff frequency parameter, the effective frequency modulation command is subjected to low-pass frequency conversion filtering and decoupled from the thermal power reference power to obtain the thermal power power command and the medium- and high-frequency residual power command. S5.3: Based on the high-pass cutoff frequency component in the cutoff frequency parameter, perform ultra-high frequency pulse stripping on the mid-to-high frequency residual power command to obtain the supercapacitor power command, and subtract the supercapacitor power command from the mid-to-high frequency residual power command to obtain the lithium battery power command.

9. The frequency regulation method for hybrid energy storage thermal power units based on hierarchical adaptive control according to claim 6, characterized in that, Step S3.2 includes: S3.2.1: Construct a coupled penalty tensor based on the fatigue index and decay index in the two-dimensional health status coordinates; S3.2.2: The collaborative crisis index is obtained by using the coupling penalty tensor to measure the collaborative crisis index of the two-dimensional health status coordinates; S3.2.3: Based on the preset crisis tolerance threshold, the collaborative crisis index is shifted and saturated to obtain the adjustment coefficient vector.

Images