Virtual power plant dynamic optimization scheduling system based on cloud platform

Abstract

The invention relates to the technical field of power system dispatching and cloud platforms, in particular to a virtual power plant dynamic optimization dispatching system based on a cloud platform, which comprises a data acquisition unit used for acquiring asset health states; the health entropy quantification unit is used for calculating the asset health entropy of the asset health state; the data credibility quantification unit is used for calculating the data credibility of the asset health state and the theoretical health state; the failure risk evaluation unit is used for calculating a cascade failure risk index; the scheduling decision-making unit is used for generating a conventional optimization scheduling signal or a detection starting signal; the active detection execution unit issues a detection instruction to the high-risk asset subset; the response deviation analysis unit is used for calculating a response deviation ratio; the state closed-loop correction unit is used for updating the theoretical health state; when the cascade failure risk index is lower than a preset threshold value, switching to generate a conventional optimization scheduling signal; the physical survivability and the anti-fragility ability of the virtual power plant in the extreme environment are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of power system dispatching and cloud platform technology, in particular to a virtual power plant dynamic optimization dispatching system based on a cloud platform. BACKGROUND

[0002] In the current virtual power plant (VPP) operating environment, system dispatching is highly dependent on the real-time health status data reported by the aggregated DERs units; the existing optimization dispatching scheme usually takes economic benefits as the target and assumes that the reported data is reliable; However, this dispatching mechanism has serious defects: the physical health status of DERs assets deteriorates with operation; the system may suffer from malicious data injection attacks, resulting in a serious mismatch between the reported health status and the true physical capacity of the assets; when the system relies on these false data for decision-making, it will produce a ghost power risk, i.e., the promised dispatchable capacity does not exist in reality; this risk will seriously threaten the physical survivability of the VPP and the safety of the power grid when the power grid is under high load or is close to the collapse boundary; therefore, how to dynamically quantify the cascading failure risk determined by the actual health status of the assets and the data reliability, and actively obtain the true physical response of the assets to close-loop correct the system cognition when the risk surges, to ensure the anti-fragile ability of the VPP in complex environments, is a technical problem that needs to be solved. SUMMARY

[0003] To solve the above technical problems, the present application provides a virtual power plant dynamic optimization dispatching system based on a cloud platform, in particular, the technical solution of the present application comprises: a data acquisition unit for acquiring the asset health status of the distributed energy unit; a health entropy quantification unit for calculating the asset health entropy of the asset health status; a data reliability quantification unit for calculating the data reliability of the asset health status and the theoretical health status; a failure risk assessment unit for calculating the cascading failure risk index in combination with the asset health entropy and the data reliability; a dispatching decision unit for comparing the cascading failure risk index with a preset threshold to generate a regular optimization dispatching signal or a detection start signal; an active detection execution unit for issuing a detection instruction to a high-risk asset subset in response to the detection start signal; a response deviation analysis unit for acquiring the actual power response and the expected power response of the high-risk asset subset and calculating the response deviation rate; a state close-loop correction unit for correcting the data reliability and updating the theoretical health status using the response deviation rate; It is also used to recalculate the asset health entropy and cascading failure risk index based on the corrected data credibility and the updated theoretical health status, and to switch to generating a regular optimized scheduling signal when the cascading failure risk index is lower than a preset threshold.

[0004] Optionally, the process for obtaining asset health entropy is as follows: The collected asset health status is mapped to a preset discrete health range; The number of assets falling within each discrete health interval is counted to calculate the probability of an asset health status falling within each interval. Based on probability, the asset health entropy is calculated using the Shannon entropy model.

[0005] Optionally, the process for obtaining data credibility is as follows: Obtain the asset health status reported by the asset and obtain the theoretical health status independently predicted by the pre-trained physical mechanism degradation model within the system; Calculate the deviation between the asset's current health status and its theoretical health status; Substituting the bias into the exponential decay trust model and combining it with a preset credibility decay coefficient, the credibility of the data is calculated.

[0006] Optionally, the calculation process for the cascading failure risk index is as follows: Obtain asset health entropy; Obtain data credibility and calculate the system's average unreliability. The asset health entropy and the system average unreliability are linearly weighted according to preset risk contribution weights to obtain the cascading failure risk index.

[0007] Optionally, the active probing execution unit is used for: In response to the detection activation signal, the data is filtered based on its reliability; Assets with the lowest location data reliability and whose reported health status is healthy or has sufficient power are set as a subset of high-risk assets. Non-economic detection commands are generated for a subset of high-risk assets; these commands are perturbation-type charge / discharge detection commands.

[0008] Optionally, the response deviation analysis unit is used for: Within the preset detection command duration window, the actual power response of a subset of high-risk assets is collected in real time. Calculate the expected power response using a physical mechanism degradation model based on the asset health status reported by a subset of high-risk assets. By using a normalized integral deviation model, the integral deviation between the actual power response and the expected power response is calculated, and the response deviation rate is obtained.

[0009] Optionally, the response deviation rate represents the dimensionless ratio obtained by dividing the integral of the absolute deviation between the actual power response and the expected power response within the detection command duration window by the absolute integral of the expected power response within the detection command duration window.

[0010] Optionally, the state closed-loop correction unit is used for: Obtain the credibility of the data before the probe and set it as the old credibility. Substituting the response deviation rate into the exponential decay function and combining it with the preset deviation sensitivity parameter, the contribution value of the measured evidence is calculated. By combining the preset learning rate, the old confidence level and the contribution value of measured evidence are recursively filtered or exponentially moved averaged to calculate the corrected data confidence level.

[0011] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention creates a unified cascading failure risk index by integrating the dispersion of asset health status with the reliability of data reporting, which can dynamically quantify the ghost power threat caused by both equipment aging and data attacks; 2. This invention can automatically suspend routine economic scheduling and switch to active detection mode when high risk is detected; the system can accurately locate suspicious assets with the lowest credibility but falsely reported status, achieve targeted verification, and actively obtain real ground information; 3. This invention quantifies the degree of discrepancy between asset behavior and actual power response under detection commands by comparing the actual power response with the expected power response; using this discrepancy, the system can close the loop to correct the internal data credibility perception and update the theoretical health status model. 4. This invention constructs a complete scheduling mechanism of risk perception, active detection, and closed-loop correction; after eliminating uncertainty and identifying and isolating high-risk assets, the system can automatically resume normal optimization, which greatly improves the physical survivability and antifragility of the virtual power plant in extreme environments. Attached Figure Description

[0012] The present invention will be further explained below with reference to the accompanying drawings and embodiments: Figure 1 This is a structural diagram of the system of the present invention. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0014] Example 1: Please see Figure 1 A cloud-based virtual power plant dynamic optimization scheduling system includes: The data acquisition unit is used to collect the asset health status of the distributed energy unit; Health entropy quantization unit, used to calculate the asset health entropy of the asset health status; The data credibility quantification unit is used to calculate the data credibility between the asset health status and the theoretical health status. The failure risk assessment unit is used to calculate the cascading failure risk index by combining asset health entropy and data credibility. The scheduling decision unit is used to compare the cascade failure risk index with a preset threshold and generate a regular optimized scheduling signal or a detection start signal. The active detection execution unit is used to issue detection commands to a subset of high-risk assets in response to a detection initiation signal; The response deviation analysis unit is used to obtain the actual power response and expected power response of a subset of high-risk assets and calculate the response deviation rate. The state closed-loop correction unit is used to correct data reliability and update the theoretical health state using the response deviation rate; It is also used to recalculate the asset health entropy and cascading failure risk index based on the corrected data credibility and the updated theoretical health status, and to switch to generating a regular optimized scheduling signal when the cascading failure risk index is lower than a preset threshold.

[0015] This invention provides a cloud-based virtual power plant dynamic optimization scheduling system. As an anti-fragile dynamic optimization scheduling system for virtual power plants (VPPs) in complex power grid environments, the system aims to address the risk of ghost power caused by the deterioration of the health status of distributed energy resources (DERs) or data injection attacks, and ensure the physical survivability of VPPs when they approach the collapse boundary. The data acquisition unit aims to obtain the real-time operating status of each distributed energy source (DER) in the virtual power plant (VPP); in this embodiment, this unit is specifically used to collect data from the VPP. Each DERs unit collects and reports operational data, among which the most critical data is asset data. Reported health status ; The purpose of the health entropy quantification unit is to quantify the dispersion of the overall health status of the VPP asset pool, i.e., one dimension of ghost power risk; in this embodiment, this unit receives data from the data acquisition unit. indivual The value is calculated based on the information theory model to determine the asset health entropy. ; The purpose of a data reliability metric unit is to evaluate reported data. To verify the authenticity of the data and prevent spectral injection attacks; in this embodiment, the unit is for each asset Maintain and calculate a data credibility This calculation does not rely on the reported data itself, but is based on the asset reports. Theoretical health status predicted independently of the system's internal model Deviation between; The failure risk assessment unit aims to create a unified internal metric for quantifying how close the system is to the 10 / 10 failure boundary; in this embodiment, the unit integrates system-level metrics from the health entropy quantification unit. and individuals from data trust measurement units It also calculates the average unreliability and uses a custom risk model to calculate the system-level cascading failure risk index. ; The scheduling decision-making unit's core function is to switch between profit and survival modes based on risk levels; in this embodiment, this unit will... With a preset active detection trigger threshold Perform a comparison; This refers to a key custom strategy parameter that represents the internal identification boundary for the system to slide from a normal profitable operating state with a risk level of 9 / 10 to the critical point of 9.5 / 10 risk. To further clarify the logic behind setting this threshold, The value of is determined based on an offline simulation platform, calibrating the critical risk value that balances system security and economic benefits by simulating scenarios of high power grid congestion and data attacks; if If the risk is deemed controllable, the system generates a conventional optimized scheduling signal, with objectives such as maximizing the expected benefits of VPP participation in grid dispatch (e.g., A) and minimizing the tracking deviation of historical dispatch commands (e.g., B). If the risk is deemed too high and the risk of the ghost power trap increases dramatically, the system will automatically suspend the normal economic optimization and generate a detection start signal. The active detection execution unit aims to proactively acquire ground-based real-time information about suspicious assets in response to high-risk signals. In this embodiment, upon receiving the detection activation signal, the unit immediately filters and locates a subset of high-risk assets. and issue specific detection commands to that subset. ; The response deviation analysis unit aims to quantify the degree of inconsistency between the actions and statements of high-risk assets. In this embodiment, after the detection command is issued, this unit will collect the actual power response of the asset in real time. And simultaneously calculate the expected power response based on its reported data. By comparing these two, the unit calculates a response deviation rate. ; The purpose of the state-closed-loop correction unit is to utilize the measured evidence obtained from active detection. To correct system cognition and achieve an antifragile closed loop; in this embodiment, the unit utilizes the response deviation analysis unit to calculate The credibility of the data for this asset Perform closed-loop correction to obtain new At the same time, the system also utilizes actual measurements Data reverse engineering forces an update to the theoretical health status in the internal physical model. Specifically, this update is achieved through an extended Kalman filter (EKF), where the measured power... As an observation, it is used to correct based on Predicted The deviation is such that the filter's state vector contains the information that needs to be corrected. ; The state closed-loop correction unit is also used to realize the automatic switching back of the system; in this embodiment, the system uses the corrected state closed-loop correction unit. and based on the updated Recalculated Substitute into the risk model and recalculate Because uncertainty is eliminated, high-risk assets... Confirmed and significantly reduced, It will decrease; when Lower than again When this happens, the unit automatically switches and regenerates the normal optimized scheduling signal, restoring the system to the normal optimized operating state; This embodiment constructs a complete risk perception-active detection-closed-loop correction scheduling system through the collaborative work of the aforementioned units. It can not only perform conventional optimized scheduling like existing technologies, but also automatically switch to a detection mode aimed at physical survival when it detects that ghost power risks are approaching the critical point due to poor asset health or data corruption. This mode actively acquires real-world information, corrects the system's state perception, and finally automatically switches back to the conventional mode after eliminating the risk. This greatly improves the anti-fragility and operational security of virtual power plants in complex power grid environments and under network attacks.

[0016] Example 2: The process of obtaining asset health entropy is as follows: The collected asset health status is mapped to a preset discrete health range; The number of assets falling within each discrete health interval is counted to calculate the probability of an asset health status falling within each interval. Based on probability, the asset health entropy is calculated using the Shannon entropy model.

[0017] In the system of Example 1, the specific quantization process of the health entropy quantization unit is as follows: The system will collect Individual asset health status Mapped to A preset discrete health interval; This refers to the total number of intervals, which is a preset adjustable parameter. Its setting is based on historical data statistics to balance computational granularity and system overhead; for example, it can be... For example, 0-100% is divided into Two ranges, 0-10%, 10-20%; The system statistics fall within each discrete health interval The number of assets within a given range is used to calculate the asset health status within that range. probability ; The system is based on this probability The asset health entropy was calculated using the Shannon entropy model. This calculation quantifies the ghost power risk between the system's actual schedulable capacity and its promised capacity. The calculation formula is as follows: ; The asset health entropy, whose physical dimension is dimensionless or bit, is the final output of this unit; Number the discrete intervals of health status, for example, from 1 to... ; The total number of intervals, as mentioned above, is a preset adjustable parameter; For asset health status Falling in the range The probability; its source is based on the system's... indivual Calculated in real time; As a system-level indicator, a higher value indicates greater variation in the health of assets, meaning a more dispersed health status within the asset pool. For example, half of the assets might be 100% healthy, and the other half 0% healthy. The higher the value, the greater the risk of ghost power the system faces. This embodiment uses the Shannon entropy model for quantification. This provides an objective and dynamic metric to measure the overall health dispersion of a VPP asset pool; compared to traditional average or worst-case assessments, this... The indicators can more sensitively capture the overall heterogeneity of asset health status, providing key, quantitative inputs for subsequent cascading failure risk assessment.

[0018] Example 3: The process of obtaining data credibility is as follows: Obtain the asset health status reported by the asset and obtain the theoretical health status independently predicted by the pre-trained physical mechanism degradation model within the system; Calculate the deviation between the asset's current health status and its theoretical health status; Substituting the bias into the exponential decay trust model and combining it with a preset credibility decay coefficient, the credibility of the data is calculated.

[0019] In the system of Example 1, the specific quantization process of the data reliability quantification unit is as follows: This unit acquires assets. Reported asset health status The parameter is obtained from the data acquisition unit, which acquires the theoretical health status independently predicted by a pre-trained physical mechanism degradation model within the system. ; This refers to a system that does not rely on reported data, but rather independently deduces theoretical health values ​​based on historical scheduling instructions and known physical factors, such as electrochemistry and degradation mechanisms. In this embodiment, the physical mechanism degradation model is preferably a Long Short-Term Memory (LSTM) network trained based on historical operational data, such as charge / discharge cycle counts, ambient temperature, and discharge rate. This network is used to predict the expected health status in the absence of data attacks. (Unit computation) and The deviation between them, i.e. ; The unit substitutes this bias into an exponentially decaying trust model and combines it with a preset trust decay coefficient. Solve for the data credibility This model is used to evaluate Spectral Data Injection attacks. Its calculation does not rely on the reported data itself, but rather compares the deviation between the reported data and the internal model data. The calculation formula is as follows: ; For assets The data reliability; its value range is , dimensionless, is the output of this unit; For assets The reported health status; its source is the data collection unit; This refers to the theoretical health state predicted by the system's internal model; its source is the physical mechanism degradation model within the system. This is the preset credibility attenuation coefficient; its dimensions are... The reciprocal of the dimensions, for example ; Used to adjust the system's sensitivity to data deviations, its value is calibrated based on statistics of the normal sensor drift range; When reported Compared with the predictions of the internal model When the levels are highly consistent, the deviation is very small. A value close to 1 indicates reliable data; when the discrepancy increases, such as during an attack, a false health value is reported. It will decrease exponentially, rapidly approaching 0, indicating that the data is highly unreliable; This embodiment employs an exponentially decaying trust model, achieving a trust assessment that relies on the deviation between the data and the physical model's expectations, rather than the data itself. This method can effectively identify reports that appear promising but are not actually true. Ghost data attacks, which are highly sophisticated but actually defy the laws of physics, provide another crucial and independent dimension for risk assessment.

[0020] Example 4: The calculation process for the cascading failure risk index is as follows: Obtain asset health entropy; Obtain data credibility and calculate the system's average unreliability. The asset health entropy and the system average unreliability are linearly weighted according to preset risk contribution weights to obtain the cascading failure risk index.

[0021] In the system of Example 1, the failure risk assessment unit calculates the cascade failure risk index. The specific process is as follows: The unit obtains the asset health entropy calculated by the health entropy quantification unit. ; The unit obtains the data trusted measurement unit calculated by the data measurement unit. Data credibility of individual assets And combined with pre-set, asset-reflecting Asset criticality weights in VPP topology ,in Calculate the weighted average unreliability, i.e. ; The unit will use asset health entropy Based on the system's average unreliability and according to the preset risk contribution weights and By performing linear weighting, the cascading failure risk index is obtained. The risk model involves creating a unified internal indicator that must simultaneously reflect the deterioration of asset health. High and data pollution The risk of ghost power caused by low commonality is calculated using the following formula: ; It is the cascade failure risk index; its physical dimension is dimensionless, and it is the output of this unit, used for subsequent decision-making. This refers to the asset health entropy; its source is calculated from the health entropy quantification unit. For the first The data credibility of an individual asset is derived from calculations by a data credibility quantification unit. Unreliable; For assets Keyness weight; Total assets; and Assign weights to pre-defined dimensionless risks; and The data is derived from offline simulation platforms that simulate high power grid congestion and data attacks for calibration, used to balance health entropy and data credibility against overall risk. Contributions; This formula is the core decision-making basis of this invention; it measures the system-level health dispersion. Data credibility of individuals Logically linked; if any factor worsens, that is... Increase or A decrease in the mean leads to an increase in average unreliability, both of which will result in... The elevation increases, thereby triggering subsequent detection mechanisms; This embodiment uses a linear weighted risk model to classify physical health risks. and information security risks Two different risk sources are merged into a single, operable risk index. This allows the system to quantify the overall threat of ghost power based on a unified benchmark, thereby providing a clear framework for subsequent scheduling decision-making units. Thresholds provide the possibility.

[0022] Example 5: The active detection execution unit is used for: In response to the detection activation signal, the data is filtered based on its reliability; Assets with the lowest location data reliability and whose reported health status is healthy or has sufficient power are set as a subset of high-risk assets. Non-economic detection commands are generated for a subset of high-risk assets; these commands are perturbation-type charge / discharge detection commands.

[0023] In the system of Example 1, the active detection execution unit responds to the detection start signal, i.e. The specific execution process is as follows: In response to the detection activation signal, the unit immediately calculates the data reliability based on the data reliability quantification unit. Perform the screening; The purpose of screening is to determine the credibility of the data. The lowest, but its reported asset health status However, these assets are displayed as healthy or fully charged; the high-risk asset subset refers to these assets; they are the most inconsistent and suspicious assets, and the system therefore sets them as the high-risk asset subset. ; The unit is a subset of the high-risk assets. Each asset in Generate a non-economical probe command Non-economic means that the purpose of the instruction is not to generate profits for the VPP, but to verify the true physical capabilities of the asset, and therefore may temporarily sacrifice economic efficiency; in this embodiment, the detection instruction is preferably a disturbance-type charge and discharge detection instruction. This embodiment achieves targeted detection by precisely screening a high-risk subset K; instead of blindly detecting all assets, it targets only those assets with the lowest credibility but the best reported status, i.e., the assets most likely to be the target of ghost data attacks; by issuing non-economical perturbation commands, this embodiment can force suspicious assets to expose their true physical state with minimal system cost and maximum efficiency.

[0024] Example 6: The response deviation analysis unit is used for: Within the preset detection command duration window, the actual power response of a subset of high-risk assets is collected in real time. Calculate the expected power response using a physical mechanism degradation model based on the asset health status reported by a subset of high-risk assets. The integral deviation between the actual power response and the expected power response is calculated using a normalized integral deviation model, and the response deviation rate is obtained. The response deviation rate represents the dimensionless ratio obtained by dividing the integral of the absolute deviation between the actual power response and the expected power response within the detection command duration window by the absolute integral of the expected power response within the detection command duration window.

[0025] In the system of Example 1, the active detection execution unit issues a detection command. The specific analysis process of the response deviation analysis unit is as follows: The unit is within the preset detection command duration window Internally, it collects high-risk asset subsets in real time. Actual power response ; It is the power response value actually measured by the system's sensors; The unit uses the asset health status reported by the asset. This means that the data suspected of being contaminated is used to calculate a predicted power response through a degradation model of the system's internal physical mechanism or its inverse model. ; This represents the data reported by the asset. It's true, it is. The expected power response under command; The unit calculates the actual power response using a normalized integral deviation model. Compared with expected power response The integral deviation between the two values ​​is used to calculate the response deviation rate. This model obtains real-world ground information about contaminated data through a non-economic command. The response deviation rate Specifically, this is expressed as the difference between the actual power response and the expected power response within the detection command duration window. The dimensionless ratio is obtained by dividing the integral value of the absolute deviation within the time window by the integral value of the expected power response within that time window. The formula for calculating this ratio is as follows: ; The response deviation rate; its physical dimension is dimensionless, and it is the output of the unit; This is the actual measured power response; its source is the sensor's response within the time window. The collected values ​​within; This is the power response expected based on the reported contaminated data; its source is the internal model within the time window. The calculated value within; This is the preset duration window for the detection command; For a preset, very small positive number, for example This is used to prevent calculation errors when the expected response integral is 0; The numerator and denominator of this formula have the same physical dimensions: power × time = energy. It is a normalized dimensionless ratio; if assets If it has not been attacked and its reported data is true, then It will be very close ,lead to Close to 0; conversely, if assets Subject to ghost data attacks, such as real If the value is very low but the reported value is very high, it will be unable to complete the detection command, resulting in... much smaller ,at this time It will increase significantly, approaching 1; This embodiment provides a precise method for quantifying the degree of inconsistency between asset statements and actions through a normalized integral deviation model; it not only determines yes or no, but also provides a... Interval deviation rate This normalization process, by dividing by the integral of the expected response, eliminates the scaling effects caused by different power level commands, making... This will become a standardized, experimentally relevant piece of evidence that can be used for subsequent closed-loop corrections.

[0026] Example 7: The state closed-loop correction unit is used for: Obtain the credibility of the data before the probe and set it as the old credibility. Substituting the response deviation rate into the exponential decay function and combining it with the preset deviation sensitivity parameter, the contribution value of the measured evidence is calculated. By combining the preset learning rate, the old confidence level and the contribution value of measured evidence are recursively filtered or exponentially moved averaged to calculate the corrected data confidence level.

[0027] In the system of Example 1, the response deviation analysis unit calculates... Subsequently, the specific correction process of the state closed-loop correction unit is as follows: The unit obtains the credibility of the asset's data prior to detection. The value is calculated by the method in Example 3 and set as the old confidence level. The unit will calculate the response deviation rate from the response deviation analysis unit. Substitute into an exponentially decaying function And combined with a preset deviation sensitivity parameter The contribution value of the measured evidence was calculated. The unit is combined with a preset learning rate. Regarding the old credibility Contribution value of measured evidence Perform recursive filtering or exponential moving average to calculate the corrected data reliability. This update mechanism is the core of the antifragile mechanism of this invention, and its calculation formula is as follows: ; The revised credibility; its physical dimensions are dimensionless, and it will be used to replace the old one. ; This is the old confidence level before the detection; its source is the calculation results of the preceding steps. This is the measured response deviation rate; its source is the calculation result of the response deviation analysis unit. The learning rate is the preset value; its range is... Dimensionless, used to control the weighting of new and old evidence in the system, i.e., the update speed; The value is determined based on the system's sensitivity requirements in responding to new evidence; This is a preset dimensionless deviation sensitivity parameter; it is a positive real number used for adjustment. The rate at which trust levels decline is penalized is calibrated through simulation. This formula will incorporate experimental evidence. Trust in history Connect them; The term mathematically ensures that the contribution values ​​of the experimental evidence are non-linearly mapped to the range (0, 1); when Very small, consistent in word and deed, close to 0, this term is close to 1, the formula will make Maintain or enhance trust; when When something is very large, and there is inconsistency between words and actions, the exponential value of this term approaches 0, and the formula will force the new... Significantly reduced, approaching zero, achieving trust isolation for this asset; This embodiment introduces an update mechanism that combines exponential decay and recursive filtering to achieve a nonlinear, evidence-based trust reconstruction. The introduction of parameters allows the system to flexibly configure the severity of punishment for inconsistencies between words and actions, The parameters control the speed at which the system forgets historical trust and accepts new evidence; this design ensures that after the system obtains real information from the ground, it can quickly and stably isolate high-risk assets from the scheduling pool, completing an antifragile closed loop.

[0028] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A cloud platform-based virtual power plant dynamic optimization scheduling system, characterized in that, The application relates to a distributed energy unit health state management method, which comprises the following steps: a data acquisition unit is used for acquiring an asset health state of a distributed energy unit; a health entropy quantification unit is used for calculating an asset health entropy of the asset health state; a data credibility quantification unit is used for calculating data credibility of the asset health state and a theoretical health state; a failure risk assessment unit is used for combining the asset health entropy and the data credibility to calculate a cascading failure risk index; a scheduling decision unit is used for comparing the cascading failure risk index with a preset threshold value to generate a regular optimization scheduling signal or a detection starting signal; an active detection execution unit is used for issuing a detection instruction to a high-risk asset subset in response to the detection starting signal; a response deviation analysis unit is used for acquiring an actual power response and an expected power response of the high-risk asset subset and calculating a response deviation rate; a state closed-loop correction unit is used for correcting the data credibility and updating the theoretical health state by using the response deviation rate; and the method is also used for recalculating the asset health entropy and the cascading failure risk index based on the corrected data credibility and the updated theoretical health state, and switching to generate the regular optimization scheduling signal when the cascading failure risk index is lower than the preset threshold value. 2.The cloud platform-based dynamic optimization and scheduling system for virtual power plant according to claim 1, wherein, The acquisition process of the asset health entropy is as follows: the acquired asset health state is mapped to preset discrete health intervals; the number of assets falling in each discrete health interval is counted to calculate the probability of the asset health state falling in each interval; and the asset health entropy is calculated by a Shannon entropy model based on the probability. 3.The cloud platform-based dynamic optimization and scheduling system for virtual power plant according to claim 1, wherein, The acquisition process of the data credibility is as follows: the asset health state reported by the asset is acquired, and a theoretical health state independently predicted by a pre-trained physical mechanism degradation model in the system is acquired; the deviation between the asset health state and the theoretical health state is calculated; the deviation is substituted into an exponential decay trust model, and a preset credibility decay coefficient is combined to calculate the data credibility. 4.The cloud platform-based dynamic optimization and scheduling system for virtual power plant according to claim 1, wherein, The calculation process of the cascading failure risk index is as follows: the asset health entropy is acquired; the data credibility is acquired, and a system average uncredibility is calculated; the asset health entropy and the system average uncredibility are linearly weighted according to a preset risk contribution weight to obtain the cascading failure risk index. 5.The cloud platform-based dynamic optimization and scheduling system for virtual power plant according to claim 1, wherein, The active detection execution unit is used for: screening according to the data credibility in response to the detection starting signal; locating assets with the lowest data credibility and the asset health state displayed as healthy or sufficient power as the high-risk asset subset; generating a non-economic detection instruction for the high-risk asset subset, and the detection instruction is a disturbance type charging and discharging detection instruction. 6.The cloud platform-based dynamic optimization and scheduling system for virtual power plant according to claim 1, wherein, The response deviation analysis unit is used for: acquiring the actual power response of the high-risk asset subset in a preset detection instruction duration window; calculating the expected power response by a physical mechanism degradation model by using the asset health state reported by the high-risk asset subset; calculating the integral deviation between the actual power response and the expected power response by a normalized integral deviation model to calculate the response deviation rate. 7.The cloud platform-based dynamic optimization and scheduling system for virtual power plant according to claim 6, wherein, The response deviation rate represents the absolute deviation integral value of the actual power response and the expected power response in the detection instruction duration window, divided by the absolute integral value of the expected power response in the detection instruction duration window, to obtain a dimensionless ratio. 8.The cloud platform-based dynamic optimization and scheduling system for virtual power plant according to claim 1, wherein, The state closed-loop correction unit is used for: Obtaining the data reliability before detection, and setting as the old reliability; Substituting the response deviation rate into an exponential decay function, and combining a preset deviation sensitivity parameter to obtain a measured evidence contribution value; Combining a preset learning rate to recursively filter or exponentially moving average the old reliability and the measured evidence contribution value, and solving the corrected data reliability.

Images