Intelligent control method and device for low-carbon park optical storage direct-flexible system based on brain emotion learning

Abstract

The invention is suitable for the technical field of power grid intelligent control, and provides an intelligent control method and device for a low-carbon park optical storage direct-flexible system based on brain emotion learning, and the method comprises the steps: obtaining the output voltage and output current of each distributed power generation unit; calculating a global average voltage and a weighted average current; for each distributed power generation unit, generating a sensory input signal and an emotion signal; inputting the sensory input signal and the emotion signal into a brain emotion learning model to generate a compensation signal; superposing the compensation signal to a local droop control link, and correcting the reference voltage; wherein the sensory input signal represents the voltage deviation and the current distribution deviation of the system, and the emotion signal represents the instantaneous value of the voltage deviation of the system, the integral of the voltage deviation and the current distribution deviation and the self-feedback of the compensation signal. According to the method, voltage stabilization and current accurate distribution are realized through hierarchical cooperative control, the problem of controller performance deterioration caused by model mismatch is solved, and the robustness in a high-uncertainty scene is improved.

Description

Technical Field

[0001] This invention relates to the field of smart grid control technology, and in particular to a smart control method and device for a low-carbon industrial park photovoltaic-storage-DC-flexible system based on brain emotion learning. Background Technology

[0002] The photovoltaic-storage-DC-flexible system integrates distributed photovoltaics, energy storage, and DC distribution systems into a single building, providing flexible services to the external power grid. This allows for energy consumption regulation and participation in grid demand response interactions within low-carbon buildings within industrial parks. Essentially a DC system, DC microgrids have seen increasing research in modern power grids in recent years due to their simplicity, reliability, and efficiency. Many renewable energy sources, such as photovoltaic panels, possess DC characteristics. Ensuring appropriate current load distribution is a crucial control requirement for power dispatch in DC microgrids.

[0003] The high penetration rate of renewable energy presents the power grid with supply and demand coordination challenges, especially in park-level photovoltaic-storage-DC-flexible systems, which require simultaneous precise current distribution and DC voltage stability. However, existing control methods rely heavily on precise mathematical models. The strong randomness of photovoltaic, energy storage, and flexible loads can cause fixed-parameter controllers to fail, leading to model mismatch. In addition, plug-and-play devices cause real-time changes in network impedance, which can compromise the accuracy of preset models and thus affect controller performance. Summary of the Invention

[0004] This invention provides an intelligent control method for a low-carbon park photovoltaic-storage direct-flexible system based on brain emotion learning, which is used to solve the problem of low system control stability.

[0005] The first aspect of this invention provides an intelligent control method for a low-carbon park photovoltaic-storage-flexible system based on brain emotion learning, comprising: Obtain the output voltage and output current of each distributed generation unit; Calculate the global average voltage based on the output voltage of all distributed generation units; calculate the weighted average current based on the output current of all distributed generation units and the preset droop gain coefficient. For each distributed generation unit, a sensory input signal is generated based on the global average voltage, weighted average current, and the local measured current of the current unit. Based on the global average voltage, weighted average current, and historical compensation signal, an emotion signal is generated. The sensory input signals and emotional signals are input into the brain emotion learning model to generate compensation signals; wherein the brain emotion learning model includes an adversarial learning mechanism of the amygdala module and the orbitofrontal cortex module; The compensation signal is superimposed on the local droop control loop to correct the reference voltage; The sensory input signal represents the system voltage deviation and current distribution deviation, and the emotional signal represents the instantaneous value of the system voltage deviation, the integral of the voltage deviation and current distribution deviation, and the self-feedback of the compensation signal.

[0006] Furthermore, the calculation of the global average voltage based on the output voltage of all distributed generation units includes: Global average voltage The calculation formula is as follows: in: This represents the total number of distributed generation units. For the first The output voltage of each distributed generation unit.

[0007] Furthermore, the calculation of the weighted average current based on the output current of all distributed generation units and a preset droop gain coefficient includes: Weighted average current The calculation formula is as follows: in: For the first The output current of each distributed generation unit For the first The preset droop gain coefficient of each distributed generation unit The numerical value is negatively correlated with the cell capacity.

[0008] Furthermore, for each distributed generation unit, generating a sensory input signal based on the global average voltage, weighted average current, and the local measured current of the current unit includes: in: For the first Distributed generation units at time Sensory input signals, The system's rated voltage. , The preset weighting coefficients satisfy... , .

[0009] Furthermore, the generation of the emotion signal based on the global average voltage, weighted average current, and historical compensation signal includes: in: For the first Distributed generation units at time Emotional signals, The compensation signal from the previous control cycle, i.e., the historical compensation signal, is obtained through the local storage module. To control the cycle duration, , , The preset weighting coefficients satisfy... , , .

[0010] Furthermore, the brain emotion learning model includes: The amygdala module is used to receive sensory input signals and generate excitatory outputs, the intensity of which is dynamically enhanced through the first connection weights; The orbitofrontal cortex module receives sensory input signals and generates inhibitory outputs, the intensity of which is dynamically reduced through a second connection weight. The amygdala module and the orbitofrontal cortex module process sensory input signals in parallel and interact and regulate each other through a weight update mechanism. When the emotional signal exceeds the expected output of the amygdala module, the first connection weight is strengthened; when the deviation between the compensation signal and the emotional signal exceeds a threshold, the second connection weight is weakened.

[0011] Furthermore, the step of inputting the sensory input signals and emotional signals into the brain's emotion learning model to generate compensation signals includes: in: For the first The time of each distributed generation unit The compensation signal, For the first Distributed generation units at time The output signal of the amygdala module, For the first Distributed generation units at time The output signal of the orbitofrontal cortex module, For the first Distributed generation units at time The first connection weight of the amygdala module, For the first Distributed generation units at time The second connectivity weight of the orbitofrontal cortex module.

[0012] Furthermore, the local droop control mechanism includes: The local controller of each distributed generation unit performs the following operations: Measure the local output current; The initial reference voltage is calculated based on the preset droop function, wherein the initial reference voltage = system rated voltage - (local line resistance + preset droop gain coefficient) × local output current; The initial reference voltage is output to the voltage control loop to drive the power converter to adjust the output voltage.

[0013] Furthermore, the step of superimposing the compensation signal onto the local droop control loop to correct the reference voltage includes: in: For the first The time of each distributed generation unit Corrected reference voltage, For the first The local line resistance of each distributed generation unit is an inherent hardware parameter. For the first The time of each distributed generation unit The local real-time output current.

[0014] A second aspect of the present invention provides an intelligent control device for a low-carbon campus light-storage direct-flexible system based on brain emotion learning, comprising: The output voltage and current acquisition unit is used to acquire the output voltage and output current of each distributed generation unit. The global average voltage and weighted average current calculation unit is used to calculate the global average voltage based on the output voltage of all distributed generation units; and to calculate the weighted average current based on the output current of all distributed generation units and the preset droop gain coefficient. A sensory input signal generation unit is used to generate a sensory input signal for each distributed generation unit based on the global average voltage, weighted average current, and the local measured current of the current unit. An emotion signal generation unit is used to generate an emotion signal based on the global average voltage, weighted average current, and historical compensation signal. The compensation signal generation unit is used to input the sensory input signals and emotional signals into the brain emotion learning model to generate compensation signals, wherein the brain emotion learning model includes an adversarial learning mechanism of the amygdala module and the orbitofrontal cortex module. A reference voltage correction unit is used to superimpose the compensation signal onto the local droop control loop to correct the reference voltage; wherein, the sensory input signal represents the system voltage deviation and current distribution deviation, and the emotional signal represents the instantaneous value of the system voltage deviation, the integral of the voltage deviation and current distribution deviation, and the self-feedback of the compensation signal.

[0015] As can be seen from the above technical solutions, the present invention has the following advantages: This invention achieves voltage stability and precise current allocation through hierarchical collaborative control. First, the output voltage and current of each distributed generation unit are acquired locally. Second, the global average voltage and weighted average current are calculated to generate a sensory input signal that integrates voltage and current allocation deviations, and this signal is combined with historical compensation signals to construct an emotional signal. Finally, a dynamic compensation signal is generated through an adversarial learning mechanism involving the amygdala and orbitofrontal cortex, which is then superimposed on the local drooping function to correct the reference voltage. This invention's method, through a bio-inspired model-free learning mechanism, simultaneously achieves bus voltage recovery and capacity-proportional current allocation without requiring prior knowledge of system parameters and structure. This effectively solves the controller performance degradation problem caused by model mismatch in existing technologies and improves robustness in high-uncertainty scenarios. Attached Figure Description

[0016] Figure 1 This is a schematic flowchart of an embodiment of an intelligent control method for a low-carbon park photovoltaic-storage direct-flexible system based on brain emotion learning in this invention. Figure 2 This is a topology diagram of the localized measurement system architecture in this invention; Figure 3 This is a diagram showing the output voltage distribution under the simulated resistive load test conditions in this invention. Figure 4 This is a diagram showing the output current distribution of each DG under the simulated resistive load test conditions in this invention. Detailed Implementation

[0017] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “corresponding to,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0018] Example 1 The implementation method in this embodiment can be implemented in a system, on a server, or on a terminal; no specific limitation is made. The following section will describe the intelligent control method for a low-carbon park photovoltaic-storage-flexible system based on brain emotion learning, as described in this application, from the perspective of system implementation. Please refer to... Figure 1 The method provided in this application includes the following steps: 101. Obtain the output voltage and output current of each distributed generation unit; This step involves real-time acquisition of electrical parameters from each distributed generation unit (DG) through localized measurements. The physical location of the measurement points and the system architecture are as follows: Figure 2 As shown, each distributed generation unit (including photovoltaic units #1 / #3, energy storage units #2 / #5, and flexible interface unit #4) is connected to an independent LC filter and line resistor ( , , , , The connection to the DC bus and the measurement process are as follows: At the output of the LC filter of each converter (i.e., the capacitor) and Nodes between line resistors, Deploy isolated voltage sensors to directly acquire the local terminal voltage of the converter. In the output inductor - The downstream branch uses a series perforated current sensor to measure the net output current injected into the bus in real time. In addition, magnetically shielded Rogowski coils are used to suppress switching noise from adjacent converters, and voltage signals are transmitted through triple-shielded cables; each unit is synchronously sampled at a frequency of 10kHz based on GPS second pulses, and the local controller completes ADC conversion within 50μs, providing millisecond-level real-time data for the brain emotion learning model.

[0019] 102. Calculate the global average voltage based on the output voltage of all distributed generation units; calculate the weighted average current based on the output current of all distributed generation units and the preset droop gain coefficient; Specifically, the global average voltage is used to quantify the overall voltage deviation of the system, and the calculation formula is as follows: in: This represents the total number of distributed generation units. For the first The output voltage of each distributed generation unit.

[0020] The global average voltage, as the core input of the sensor input signal, is used to calculate the voltage deviation term, which reflects the global health status of the bus voltage.

[0021] The weighted average current is used to assess the fairness of current allocation, and the calculation formula is as follows: in: For the first The output current of each distributed generation unit For the first The preset droop gain coefficient of each distributed generation unit The numerical value is negatively correlated with the cell capacity.

[0022] The weighted average current generates the current distribution deviation term of the sensor input signal, and the load ratio distribution accuracy is dynamically verified.

[0023] 103. For each distributed generation unit, a sensory input signal is generated based on the global average voltage, weighted average current, and the local measured current of the current unit; The sensory input signal mimics the biological sensory system, converting system deviations into neural excitation signals. It is used to fuse the overall voltage deviation and local current distribution deviation to form the initial input for control decisions. As the primary sensing layer of BELBIC, it triggers the generation of compensation signals. Its calculation formula is as follows: in: For the first Distributed generation units at time Sensory input signals, The system's rated voltage. , The preset weighting coefficients satisfy... , .

[0024] 104. Generate emotion signals based on global average voltage, weighted average current, and historical compensation signals; Emotional signals are a mechanism that simulates emotional responses to assess control effectiveness and self-correct. The calculation formula is as follows: in: For the first Distributed generation units at time Emotional signals, The compensation signal from the previous control cycle, i.e., the historical compensation signal, is obtained through the local storage module. To control the cycle duration, , , The preset weighting coefficients satisfy... , , It should be noted that The item is to punish the current deviation. The item is the cumulative historical bias. This section is through... and Difference adjustment of amygdala Weights are used to enable reward-driven learning.

[0025] 105. Sensory input signals and emotional signals are input into the brain emotion learning model to generate compensation signals; the brain emotion learning model includes adversarial learning mechanisms of the amygdala module and the orbitofrontal cortex module; The brain emotion learning model in this step includes: The amygdala module is used to receive sensory input signals and generate excitatory outputs, the intensity of which is dynamically enhanced through the first connection weights; The orbitofrontal cortex module receives sensory input signals and generates inhibitory outputs, the intensity of which is dynamically reduced through a second connection weight. The amygdala module and the orbitofrontal cortex module process sensory input signals in parallel and interact and regulate each other through a weight update mechanism. When the emotional signal exceeds the expected output of the amygdala module, the first connection weight is strengthened; when the deviation between the compensation signal and the emotional signal exceeds a threshold, the second connection weight is weakened.

[0026] Specifically, the implementation principle of the brain emotion learning model is as follows: the amygdala module (excitatory pathway) and the orbitofrontal cortex module (inhibitory pathway) dynamically balance and compensate for the output through antagonistic weight updates, thereby achieving model-free adaptive control.

[0027] The formula for calculating the compensation signal is as follows: in: For the first The time of each distributed generation unit The compensation signal, For the first Distributed generation units at time The output signal of the amygdala module, For the first Distributed generation units at time The output signal of the orbitofrontal cortex module, For the first Distributed generation units at time The first connection weight of the amygdala module, For the first Distributed generation units at time The second connectivity weight of the orbitofrontal cortex module; The evolution is carried out according to the following integral equation, where This is a shorthand form for variables without subscripts, and is practically equivalent to... (Each unit is calculated independently); similarly , , , For abbreviated form: , and , The instantaneous rate of change of the connection weights, The global learning rate, , The initial values ​​for the connection weights are... These are neurons that receive the maximum sensory signals directly from the thalamus. For thalamic connectivity weights (specific neuron gains). The dynamic process is as follows: , The instantaneous rate of change of thalamic connectivity weights. Let these be the initial values ​​for the thalamic connection weights; assuming... When using only one sense, Evolved into .

[0028] The evolution is carried out according to the following integral equation: ,at the same time , The instantaneous rate of change of the second connection weight. Let these be the initial values ​​for the second connection weights. The preset inhibition rate, .

[0029] 106. The compensation signal is superimposed on the local droop control loop to correct the reference voltage; wherein, the sensory input signal represents the system voltage deviation and current distribution deviation, and the emotional signal represents the instantaneous value of the system voltage deviation, the integral of the voltage deviation and current distribution deviation, and the self-feedback of the compensation signal.

[0030] This step involves local droop control, including: The local controller of each distributed generation unit performs the following operations: Measure the local output current; The initial reference voltage is calculated based on the preset droop function. The initial reference voltage = system rated voltage - (local line resistance + preset droop gain coefficient) × local output current. The initial reference voltage is output to the voltage control loop to drive the power converter to adjust the output voltage.

[0031] Specifically, the local droop control mechanism achieves primary power distribution through localized autonomous adjustment: each distributed generation unit independently measures its output current. According to the preset droop function The initial reference voltage is calculated and output to the voltage loop PI controller to drive the power converter. Its core function is to: Differential current distribution is achieved by sharing the current according to the capacity ratio, such as It has no communication dependency and a response delay of <10ms during load switching; it provides an initial reference voltage base for the secondary control layer to compensate for voltage drops.

[0032] The corrected formula for calculating the reference voltage is as follows: in: For the first The time of each distributed generation unit Corrected reference voltage, For the first The local line resistance of each distributed generation unit is an inherent hardware parameter. For the first The time of each distributed generation unit The local real-time output current.

[0033] The following is based on a MATLAB / Simulink environment and is called as follows: Figure 2 The research model shown consists of five DC / DC converters (#1-#5) connected in parallel to the DC bus, a 120V DC power supply, and an LC filter ( =0.1H, =2200μF, ), line resistance (0.01Ω~0.05Ω), droop gain , (Capacity ratio 1:2), BELBIC parameters , After optimization and adjustment.

[0034] Under the above parameters and model, a resistive load test was conducted, mainly examining the control effect of the above embodiment under a resistive load. It is assumed that the system went through the following four stages: Stage 1 (0-2s): 20Ω load was connected, with only droop control; Stage 2 (2-4s): secondary control activated BELBIC; Stage 3 (4-6s): the load increased to 40Ω; Stage 4 (6-8s): the load was removed, and the stability was verified.

[0035] The simulation results of the output voltage and output current of each DG in this study example are as follows: Figure 3 and Figure 4 As shown. From Figure 3 As can be seen, in the first two seconds using only local droop control, the DC bus voltage will drop to approximately 47.5V, deviating from the expected value of 48V. The secondary controller in... After 2 seconds of activation, the bus voltage quickly recovers to approximately 48V. Furthermore, when the load connected to the bus changes, the bus voltage remains well maintained around the nominal value. Figure 4 As can be seen, regardless of load changes, the method in the above embodiments can ensure appropriate current sharing. The analysis results of this example demonstrate that the control method proposed in this invention guarantees both proportional current distribution and normal voltage recovery.

[0036] Example 2 An embodiment of the intelligent control device for a low-carbon park photovoltaic-storage-flexible system based on brain emotion learning in this invention includes the following: The output voltage and current acquisition unit is used to acquire the output voltage and output current of each distributed generation unit. The global average voltage and weighted average current calculation unit is used to calculate the global average voltage based on the output voltage of all distributed generation units; and to calculate the weighted average current based on the output current of all distributed generation units and the preset droop gain coefficient. The sensory input signal generation unit is used to generate a sensory input signal for each distributed generation unit based on the global average voltage, the weighted average current, and the local measured current of the current unit. The emotion signal generation unit is used to generate emotion signals based on global average voltage, weighted average current and historical compensation signals; The compensation signal generation unit is used to input sensory input signals and emotional signals into the brain emotion learning model to generate compensation signals. The brain emotion learning model includes an adversarial learning mechanism of the amygdala module and the orbitofrontal cortex module. The reference voltage correction unit is used to superimpose the compensation signal onto the local droop control loop to correct the reference voltage. Among them, the sensory input signal represents the system voltage deviation and current distribution deviation, and the emotional signal represents the instantaneous value of the system voltage deviation, the integral of the voltage deviation and current distribution deviation, and the self-feedback of the compensation signal.

[0037] It is understood that those skilled in the art can combine various implementation methods in the above embodiments under the guidance of the above examples to obtain technical solutions with multiple implementation methods.

[0038] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A low-carbon park light storage direct flexible system intelligent control method based on brain emotional learning, characterized by, The method comprises: acquiring output voltage and output current of each distributed power generation unit; calculating global average voltage according to output voltage of all distributed power generation units; calculating weighted average current according to output current of all distributed power generation units and preset droop gain coefficient; generating sensory input signal based on the global average voltage, weighted average current and local measured current of the current unit for each distributed power generation unit; generating emotion signal based on the global average voltage, weighted average current and historical compensation signal; inputting the sensory input signal and emotion signal into a brain emotion learning model to generate compensation signal, wherein the brain emotion learning model comprises an amygdala module and an orbitofrontal cortex module in an antagonistic learning mechanism; superimposing the compensation signal into a local droop control link to correct reference voltage; wherein the sensory input signal represents system voltage deviation and current allocation deviation, and the emotion signal represents system voltage deviation instantaneous value, integral of voltage deviation and current allocation deviation and self-feedback of compensation signal. 2.The low-carbon park light storage direct flexible system intelligent control method based on brain emotional learning of claim 1, wherein, The calculation of the global average voltage according to the output voltage of all distributed power generation units comprises: Global average voltage The formula for the calculation is: wherein: is the total number of distributed generation units, is the output voltage of the th distributed generation unit. 3.The intelligent control method of low-carbon park light storage direct flexible system based on brain emotional learning according to claim 2, characterized in that, The calculation of the weighted average current according to the output current of all distributed power generation units and the preset droop gain coefficient comprises: Weighted average current The formula for the calculation: wherein: is the output current of the th distributed generation unit, is the preset droop gain coefficient of the th distributed generation unit, the numerical value is negatively correlated with the unit capacity. 4.The intelligent control method of low-carbon park light storage direct flexible system based on brain emotional learning according to claim 1, characterized in that, The generation of the sensory input signal based on the global average voltage, weighted average current and local measured current of the current unit for each distributed power generation unit comprises: wherein: is the number of the th distributed power generation unit, is the sensory input signal of the th distributed power generation unit at time , is the system rated voltage, , is a preset weight coefficient, satisfying 5.The intelligent control method of low-carbon park light storage direct flexible system based on brain emotional learning according to claim 4, characterized in that, The generation of the emotion signal based on the global average voltage, weighted average current and historical compensation signal comprises: Wherein: is the first is the second is the mood signal of the is the compensation signal of the last control cycle, i.e. the historical compensation signal, obtained through a local storage module, is the control cycle length, , , is a preset weight coefficient, satisfying , , .

6. The low-carbon park light storage direct flexible system intelligent control method based on brain emotional learning according to any one of claims 1-5, characterized in that, The brain emotion learning model comprises: The amygdala module is configured to receive the sensory input signal and generate excitatory output, and the output intensity is dynamically enhanced through a first connection weight; The orbitofrontal cortex module is configured to receive the sensory input signal and generate inhibitory output, and the output intensity is dynamically weakened through a second connection weight; The amygdala module and the orbitofrontal cortex module process the sensory input signal in parallel and interactively adjust through a weight updating mechanism; when the emotion signal exceeds the expected output of the amygdala module, the first connection weight is enhanced; when the deviation between the compensation signal and the emotion signal exceeds a threshold, the second connection weight is weakened. 7.The intelligent control method of the low-carbon park light storage direct flexible system based on brain emotional learning according to claim 6, characterized in that, The inputting of the sensory input signal and emotion signal into the brain emotion learning model to generate the compensation signal comprises: wherein: is an output signal of the amygdala module of the jth distributed power generation unit at time is a compensation signal for the jth distributed power generation unit at time is an output signal of the orbitofrontal cortex module of the jth distributed power generation unit at time is an output signal of the amygdala module of the jth distributed power generation unit at time is an output signal of the orbitofrontal cortex module of the jth distributed power generation unit at time is a first connection weight of the amygdala module of the jth distributed power generation unit at time is a second connection weight of the orbitofrontal cortex module of the jth distributed power generation unit at time ​​​​​​​​ 8.The intelligent control method of low-carbon park light storage direct flexible system based on brain emotional learning according to claim 1, characterized in that, The local droop control link comprises: The local controller of each distributed power generation unit performs the following operations: measuring local output current; calculating initial reference voltage according to a preset droop function, the initial reference voltage = system rated voltage - (local line resistance + preset droop gain coefficient) × local output current; outputting the initial reference voltage to a voltage control loop to drive the power converter to adjust the output voltage. 9.The intelligent control method of the low-carbon park light storage direct flexible system based on brain emotional learning according to claim 8, characterized in that, The superimposition of the compensation signal into the local droop control link to correct the reference voltage comprises: wherein: is the modified reference voltage of the th distributed generation unit at time is the local line resistance of the th distributed generation unit, which is a hardware inherent parameter, is the local real-time output current of the th distributed generation unit at time .​ 10. A low-carbon park light storage direct flexible system intelligent control device based on brain emotional learning, characterized by The method according to any one of claims 1-9 comprises: an output voltage and current acquisition unit configured to acquire output voltage and output current of each distributed power generation unit; a global average voltage and weighted average current calculation unit configured to calculate a global average voltage according to output voltages of all the distributed power generation units, and calculate a weighted average current according to output currents of all the distributed power generation units and a preset droop gain coefficient; a sensory input signal generation unit configured to generate, for each distributed power generation unit, a sensory input signal based on the global average voltage, the weighted average current and a locally measured current of the current unit; an emotion signal generation unit configured to generate an emotion signal based on the global average voltage, the weighted average current and a historical compensation signal; a compensation signal generation unit configured to input the sensory input signal and the emotion signal into a brain emotion learning model to generate a compensation signal, wherein the brain emotion learning model comprises an antagonistic learning mechanism of an amygdala module and an orbitofrontal cortex module; a reference voltage correction unit configured to superimpose the compensation signal into a local droop control link to correct a reference voltage; wherein the sensory input signal represents a system voltage deviation and a current allocation deviation, and the emotion signal represents a system voltage deviation instantaneous value, an integral of the voltage deviation and the current allocation deviation and a self-feedback of the compensation signal.

Images