A data-driven based new energy storage power station collaborative operation management and control system

Abstract

This invention relates to the field of new energy storage power station operation and management technology, and discloses a data-driven collaborative operation and management system for new energy storage power stations. The system includes: a system modeling and state perception module that uses a stochastic hybrid automaton to model the cloud-edge collaborative system of the energy storage power station in an integrated manner. This module characterizes the continuous dynamics and random disturbances of the equipment's physical evolution and constructs a stochastic transfer model of discrete events in cloud-edge interaction. Combined with the system state definition of multiple operating modes and switching conditions, it acquires the real-time state and completes state estimation. An information value decision-making and primary triggering module determines the state based on the state estimation results and local information sets. This data-driven collaborative operation and management system for new energy storage power stations addresses the core defects of existing technologies, such as poor adaptability to complex operating conditions, weak module collaboration, and an imbalance between safety and value. It avoids ineffective communication and energy waste, ultimately achieving a dynamic balance between the safety of the energy storage power station and its economical operation, significantly improving the system's operational reliability and overall benefits.

Description

Technical Field

[0001] This invention relates to the field of operation and management technology for new energy storage power stations, and in particular to a data-driven collaborative operation and management system for new energy storage power stations. Background Technology

[0002] Current new energy storage power stations achieve basic operation and management through status perception, cloud-edge communication scheduling, and safety monitoring modules. Conventional solutions rely on fixed model parameters and empirical thresholds for judgment, and combine simple quantitative methods to try to balance economic optimization and safe operation. They are widely used in new energy consumption scenarios such as photovoltaic and wind power supporting energy storage.

[0003] Existing technologies suffer from two major flaws, making it difficult to adapt to the complex operational needs of energy storage power stations: First, they have weak adaptability to complex operating conditions such as sudden disturbances in photovoltaic systems, cloud-edge communication interruptions, and partial equipment failures. The model parameters are mostly statically set, and the independent operation of each control module lacks closed-loop linkage, making it impossible to dynamically match changes in operating conditions. Second, quantitative control is disconnected from practical implementation. Either it relies too much on theoretical calculations, resulting in poor on-site executability, or it relies solely on fixed thresholds, leading to insufficient control precision. It is impossible to achieve optimal resource allocation while ensuring system safety, which can easily lead to uncontrolled safety risks or waste of communication and energy resources. Summary of the Invention

[0004] The technical problem to be solved by this invention is that the existing technology has the following shortcomings: poor adaptability to complex working conditions, weak module coordination, and a core defect of safety and value imbalance. To this end, we propose a data-driven collaborative operation and management system for new energy storage power stations.

[0005] To achieve the above objectives, this application adopts the following technical solution: a data-driven collaborative operation and management system for new energy storage power stations, comprising: a system modeling and state perception module that integrates cloud-edge collaborative system modeling of energy storage power stations based on stochastic hybrid automata, characterizing the continuous dynamics and random disturbances of equipment physical evolution, and constructing a stochastic transfer model of discrete events in cloud-edge interaction, combined with system state definitions of multiple operating modes and switching conditions, to obtain real-time state and complete state estimation; an information value decision-making and primary triggering module that, based on the state estimation results and a local information set containing historical control commands, equipment operating history, and recent disturbance characteristic data, constructs a dynamic threshold, wherein the threshold is formed by coupling the communication cost corresponding to the current network condition and the safety coefficient adapted to the current operating mode, thereby integrating information value with dynamic... Threshold comparison generates a primary communication command that will trigger or not trigger. The probabilistic safety monitoring and risk assessment module defines a safety set for the system's critical safety states. Based on a stochastic system model, it calculates the lower bound of the probability that the system state will remain in the safety set within a preset time domain in the future, constructs a probabilistically guaranteed safety evolution boundary, and calculates the instantaneous risk probability of the current state trajectory violating this boundary in real time online. The safety arbitration and execution module is used to make the final decision based on the instantaneous risk probability and the primary communication command and execute the corresponding control strategy. When the abnormal working condition adaptive module detects a network interruption, it suppresses communication triggering and switches to an edge autonomous mode that depends on the probabilistically guaranteed safety evolution boundary. When the network recovers or an abnormal device is detected, it dynamically adjusts the operating mode and parameters, quickly synchronizes the state, and restores collaborative operation or adapts to abnormal working condition management.

[0006] Preferably, the system modeling and state awareness module describes the continuous dynamics and random disturbances of the physical evolution of the device through stochastic differential equations, and constructs a stochastic transition model of cloud-edge interactive discrete events using Markov decision processes. The multiple operating modes include normal mode, optimization mode and emergency mode, and the switching conditions are determined based on stochastic stability theory and the correlation between system state boundaries.

[0007] Preferably, the online calculation of the information value is based on solving a stochastic optimal control problem, and its value is equal to the difference between the expected optimal cost of the system in the scenario of obtaining incremental information through triggered communication and the expected optimal cost of the system in the scenario of relying only on local information; the safety coefficient in the dynamic threshold is dynamically adjusted with the operating mode, and its value in the emergency mode is lower than its value in the normal mode and the optimization mode.

[0008] Preferably, the probabilistically guaranteed safety evolution boundary is constructed using a backtracking time-domain stochastic control framework, the safety set includes a voltage safety domain, a state-of-charge safety window, and a temperature threshold range, and the instantaneous risk probability is calculated online using an importance sampling fast approximation algorithm.

[0009] Preferably, the decision logic of the security arbitration and execution module is as follows: when the immediate risk probability is lower than the preset security risk threshold, the edge autonomy or cloud-edge communication and cloud optimization command response is executed according to the primary communication command; when the immediate risk probability reaches or exceeds the preset threshold, the local emergency control strategy is forcibly activated and a high-risk alarm is sent to the cloud, and this decision has the highest priority.

[0010] Preferably, in the abnormal operating condition adaptive module, the edge autonomous mode relies on the probability-guaranteed security evolution boundary and the local emergency strategy library to operate when the network is interrupted; when the equipment is abnormal, the system automatically switches to emergency mode, and synchronously adjusts the safety factor and risk probability calculation cycle to adapt to the special control requirements of fault conditions.

[0011] Preferably, the abnormal operating condition adaptive module detects local equipment faults by monitoring the residual sequence between the system state estimate and the sensor measured value; when the statistical characteristics of the residual sequence deviate from its historical normal fluctuation range for a preset period, it determines that the corresponding equipment has a local fault and triggers a system response.

[0012] Preferably, the probability tube constructed by the probability security monitoring and risk assessment module adopts an online rolling update mechanism. The update trigger conditions include the deviation between the current state estimate and the boundary of the probability tube exceeding a preset threshold, and the random disturbance intensity undergoing a step change. The update cycle is consistent with the system control cycle.

[0013] Preferably, the information value decision and the system's expected optimal cost function in the primary triggering module... It is composed of the weighted cost of charging and discharging loss, the cost of penalty for deviation from the safety boundary, and the cost of communication resource consumption. The weight of each cost item is dynamically adjusted based on the current operating mode, with the cost of penalty for deviation from the safety boundary having the highest weight in emergency mode.

[0014] Preferably, the state vector of the system modeling and state perception module can also be supplemented with adaptation parameters according to the type of energy storage power station. For lithium-ion battery energy storage power stations, battery cell consistency parameters are supplemented, and for flow battery energy storage power stations, electrolyte concentration parameters are supplemented, in order to adapt to the physical characteristics of different types of energy storage devices.

[0015] The technical effects and advantages of this invention are as follows: This invention addresses the core defects of existing technologies, such as poor adaptability to complex operating conditions, weak module coordination, and an imbalance between safety and value. Through adaptive adjustment of module parameters and closed-loop linkage throughout the entire process, it can accurately respond to various typical operating conditions, such as sudden disturbances in photovoltaic systems, network interruptions, and equipment failures, and dynamically match changes in operating conditions to ensure the safe and stable operation of the system under all operating conditions. At the same time, it achieves a deep integration of quantitative control logic and on-site operation. On the basis of building a solid safety bottom line through precise risk control, it optimizes cloud-edge communication and charging and discharging resource allocation, avoids ineffective communication and energy waste, and ultimately achieves a dynamic balance between the safety of energy storage power stations and economic operation, significantly improving the reliability and overall benefits of system operation. Attached Figure Description

[0016] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts:

[0017] Figure 1 This is a schematic diagram of the stochastic hybrid automata modeling structure of the present invention; Figure 2 This is a complete flowchart of the online collaborative work and secure arbitration process of this invention; Figure 3 This is a logic diagram for the security arbitration and execution decision-making of this invention. Detailed Implementation

[0018] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0019] Reference Figure 1-3 As shown, the present invention provides a technical solution: a data-driven collaborative operation and management system for new energy storage power stations, characterized in that it includes: a system modeling and state perception module, an information value decision-making and primary triggering module, a probabilistic safety monitoring and risk assessment module, a safety arbitration and execution module, and an abnormal operating condition adaptive module.

[0020] The system modeling and state awareness module performs unified modeling of the collaborative operation and management system of new energy storage power stations based on stochastic hybrid automata. The modeling process includes: characterizing the continuous dynamics and random disturbances of the physical evolution of system equipment through stochastic differential equations; establishing a stochastic transition model of discrete events in cloud-edge interaction using Markov decision processes; defining multiple operating modes and their switching conditions based on the system state; and synchronously acquiring the real-time state of the system and completing state estimation.

[0021] In continuous dynamic modeling, stochastic differential equations are used to characterize the physical evolution and random disturbance effects of core equipment such as energy storage converters, battery packs, and DC buses, and to clarify the system state vector. .in The battery is in its state of charge. This is the DC bus voltage. For output power, To determine the temperature of a single battery cell and the coupling relationship between the deterministic dynamic evolution and random perturbations of the quantification equipment, stochastic differential equations are used as the core modeling tool. The specific formulas are as follows: The above formula is the core mathematical expression for continuous dynamic modeling. The physical meaning and determination method of each parameter directly affect the model accuracy. These are deterministic dynamic terms, reflecting the inherent operating characteristics of the equipment, namely battery charging and discharging dynamics and converter power conversion laws. To control the input vector, namely the charge / discharge power command and the converter modulation signal, The diffusion term coefficient matrix represents the intensity of renewable energy output fluctuations, measurement noise, and random disturbances caused by load variations. This is a standard Wiener process, describing the probabilistic evolution of random perturbations. Model parameters. The corresponding state matrix, input matrix The coefficients are systematically identified through historical operating data and determined using the least squares method or the maximum likelihood estimation method to ensure that the model is consistent with the actual equipment characteristics.

[0022] After completing the continuous dynamic modeling at the device level, the discrete interactive behaviors in the cloud-edge collaboration process, through communication requests, command issuance, and mode switching, also need to be accurately depicted in order to achieve integrated modeling coverage of the entire system. Therefore, we enter the discrete event modeling stage.

[0023] In discrete event modeling, a stochastic transition model for cloud-edge interactive discrete events is constructed using Markov decision processes. The discrete event state space is defined, including states such as communication standby, communication trigger, command issuance in progress, and mode switching in progress; the action space covers discrete actions such as edge nodes initiating communication requests, cloud-based optimization commands being issued, and system execution of mode switching. Based on historical cloud-edge communication logs and device control command execution records, the transition probabilities between each discrete state are statistically analyzed to form a stochastic transition matrix, accurately characterizing the uncertain evolution of discrete events.

[0024] Modeling continuous dynamics and discrete events provides a basic evolutionary framework for the system. However, in actual operation, the system needs to adapt to different safety and optimization requirements. Therefore, it is necessary to define differentiated operating modes and switching rules to achieve dynamic adaptation between the model and operating conditions. Based on stochastic Lyapunov stability theory, and combined with the system's safe operating boundary and optimization objective, three core operating modes are defined: normal mode, economic optimization mode, and emergency mode. The state boundary conditions of each mode are clearly defined.

[0025] For example, in normal mode, the state of charge, voltage, and temperature are all within the normal safe range; in optimized mode, the state variables are in a better operating range to pursue economy; and in emergency mode, the state variables are close to or exceed the safety threshold. The trigger logic for mode switching is formulated. When the estimated value of the system state meets the preset boundary conditions and is maintained for a set time, the mode is randomly switched. The switching probability is determined by the degree of deviation of the current state and the random stability index.

[0026] The effective implementation of the operating mode relies on high-precision real-time status data. Therefore, through a professional perception and estimation process, the physical equipment status is transformed into quantitative data that can be identified by the model and used for decision-making. Real-time operating data is collected by voltage sensors, current sensors, temperature sensors, and state of charge detectors deployed on the equipment, with the data sampling frequency adapted to the control cycle. The collected data is preprocessed, including outlier removal and noise reduction using moving average filtering. Based on the constructed stochastic hybrid automata model, the extended Kalman filter algorithm is used for state estimation. By fusing sensor data and model predictions, the optimal state estimation result is output, providing high-precision input for information value calculation and risk assessment.

[0027] The system modeling and state awareness module provides core data support for subsequent communication decisions by outputting high-precision state estimation results through integrated modeling. The information value decision and primary triggering module, as the intelligent decision-making center of the system, quantifies the actual benefits of communication based on this state data, accurately judges the necessity of communication, and avoids resource waste and omission of key communication caused by ineffective communication.

[0028] The information value decision and primary triggering module calculates the information value of cloud-edge communication at this moment online based on the state estimation results and a local information set containing historical control commands, equipment operation history, and recent disturbance characteristic data. It constructs a dynamic threshold, which is determined by the communication cost of the current network condition coupled with the security coefficient adapted to the current operating mode. The information value is compared with the dynamic threshold to generate a primary communication trigger command to trigger or not trigger communication. The specific implementation process is as follows.

[0029] The system modeling and state awareness module outputs high-precision state estimation results through integrated modeling. It provides core data support for communication decisions. The information value decision and primary trigger module serves as the intelligent decision-making center of the system. Based on this state data, it quantifies the actual benefits of communication, accurately judges the necessity of communication, and avoids resource waste and omission of key communication caused by ineffective communication. Its specific working process is as follows.

[0030] The information value decision and primary triggering module calculates the information value of cloud-edge communication at this moment online based on the state estimation results and a local information set containing historical control commands, equipment operation history, and recent disturbance characteristic data. It constructs a dynamic threshold, which is determined by the communication cost of the current network condition coupled with the security coefficient adapted to the current operating mode. The information value is compared with the dynamic threshold to generate a primary communication trigger command to trigger or not trigger communication. The specific implementation process is as follows.

[0031] Online calculation of information value based on state estimation results As the core input, it deeply integrates local information. It includes data such as historical control command execution effects, equipment operation history, recent random disturbance characteristics, and local policy execution records. By quantifying the performance difference between acquiring additional information and relying solely on local information, it calculates the value of the information. Its core logic is: comparing the incremental information acquired in the cloud after triggering communication. And achieve global optimization and rely only on local information. The system expects optimal performance in two scenarios when implementing local policies, and the difference between the two is the information value.

[0032] The information value measurement formula is used as the core calculation tool. The specific formula is as follows: The physical meaning and computational logic of each parameter in the above formula directly determine the accuracy of the information value. To determine the system's expected optimal cost function, the system comprehensively considers charging and discharging losses and the risk of deviating from the safety boundary. The mathematical expectation operator is used to characterize average performance under random conditions. This indicates a scenario where the cloud obtains complete information after the communication is triggered. This represents a scenario that relies solely on local information. The computation employs a linear quadratic Gaussian approximation method to simplify the solution complexity, ensuring real-time response within the control cycle and meeting online decision-making requirements.

[0033] The construction of dynamic thresholds and the determination of information value need to adapt to real-time operating conditions. Fixed thresholds cannot cope with changes in demand caused by network fluctuations and switching of operating modes. Therefore, it is necessary to construct dynamic thresholds that dynamically match operating conditions. This dynamic threshold is generated through a coupled logic of communication cost and mode adaptation safety factor, taking into account both resource consumption and scenario requirements: communication cost Based on real-time network status modeling, and integrating key indicators such as available bandwidth, communication latency, and packet loss rate, the security factor is calculated online through a parameterized model. The system dynamically adjusts its values ​​based on its current operating mode. In emergency mode, the values ​​are lower to prioritize the transmission of critical information, while in normal and optimized modes, the values ​​are higher to suppress unnecessary communication. Ultimately, a dynamic threshold adapted to the current operating condition is formed through linear coupling. .

[0034] After completing the information value calculation and dynamic threshold construction, it is necessary to clarify the necessity of communication through quantitative comparison and set clear judgment rules: when the real-time information value Greater than the dynamic threshold When the communication benefit exceeds the resource cost, a primary instruction to trigger communication is generated; when the information value is less than or equal to a dynamic threshold, the communication benefit is determined to be insufficient to cover the cost, and a primary instruction not to trigger communication is generated. This primary instruction is synchronously transmitted to the security arbitration and execution module, providing a basis for the system's final decision and ensuring the accuracy and timeliness of communication decisions.

[0035] The basic communication commands generated by the information value decision-making and primary triggering modules only focus on the matching of communication benefits and costs, without fully considering the security risks brought about by the random evolution of the system. As the security core of the system, the probabilistic security monitoring and risk assessment module inherits the stochastic system model formed by the preceding integrated modeling. By quantifying the secure accessibility of the system state and assessing the risk level in real time, it provides a rigid basis for security arbitration, ensuring that value-driven communication decisions do not deviate from the security bottom line.

[0036] The probabilistic safety monitoring and risk assessment module, based on the stochastic system model formed by the integrated modeling, calculates the lower bound of the probability that the system state will remain within the safety set in the future preset time domain, and constructs a probabilistically guaranteed safety evolution boundary; it calculates the instantaneous risk probability of the current state trajectory violating the probabilistic constraints online in real time. The specific implementation process is as follows.

[0037] To clarify the boundaries of safe system operation, a mathematical safety set is defined for the core safety indicators of energy storage power stations. This set covers the safety ranges of critical states, including but not limited to the voltage safety domain of the allowable fluctuation range of DC bus voltage, the safety window of the state of charge (SOC) range to avoid overcharging and over-discharging, and the maximum temperature limit to prevent thermal runaway.

[0038] Defining only a safe set is insufficient to address state fluctuations caused by random disturbances. It is necessary to quantify the probability that the system state will remain in the safe set for a period of time in the future and construct a probabilistically guaranteed safe evolution boundary with a clear confidence level, i.e., a probabilistic tube.

[0039] Based on a stochastic system model, calculate the future preset time domain. Internal, system state trajectory Always remain in the safe set The probability lower bound within this bound forms the probability-guaranteed safety evolution boundary.

[0040] The core formula for probabilistic reachability is as follows: The physical meaning and constraint logic of each parameter in the above formula directly determine the reliability of the safety boundary: where... To control the feasible input domain, the range of values ​​for the control strategy is limited. A probability operator, representing the likelihood of an event occurring. This is the estimated value of the current state. The lower bound of the probability is a preset value, representing the confidence level of the probability-guaranteed safe evolution boundary. This formula is solved using a back-to-time stochastic control framework, with initial probability parameters pre-calculated offline and the boundary updated online to ensure adaptation to real-time system state changes.

[0041] After constructing the probabilistically guaranteed safety evolution boundary, it is necessary to monitor the risk of the current state trajectory deviating from this boundary in real time. This should be based on the latest state estimation results. The importance sampling fast approximation algorithm is used to calculate in real time the system state trajectory within a finite number of future steps, which violates the probability-guaranteed safety evolution boundary, i.e., exceeds the safety set. Instant risk probability This algorithm significantly improves computational efficiency by focusing sampling on high-risk areas, ensuring that results are output within a control period of 10-100ms, providing real-time and accurate risk quantification data for security arbitration.

[0042] The real-time risk probability output by the probabilistic security monitoring and risk assessment module Together with the initial communication trigger commands generated by the information value decision-making module, they constitute the core input for the system's final decision. The security arbitration and execution module, as the decision-making center and execution terminal of the control system, ensures the system's security baseline while also considering the efficiency of communication resource utilization.

[0043] The security arbitration and enforcement module is based on the real-time risk probability and the preset security risk threshold. The comparison results, combined with the primary communication trigger instruction, are used to make the final decision. Through clear priority rules and execution procedures, the organic unity of security and operational efficiency is achieved. The specific implementation process is as follows.

[0044] To avoid conflicting decisions due to multiple inputs, security risk is prioritized over communication value, i.e., a preset security risk threshold is established. When the instantaneous risk probability reaches or exceeds the preset security risk threshold It directly triggers the highest level of emergency response, ignoring basic communication trigger commands; it only executes edge autonomy or cloud-edge collaboration based on communication value decisions when the system is in a low-risk zone.

[0045] Decision-making and execution in low-risk scenarios, based on the immediate probability of risk. At this time, the system is within a safe and controllable range, and decisions are guided by communication value, adapting to the primary communication trigger command to execute differentiated operations.

[0046] If the primary communication trigger command is not to trigger communication, it is determined that the current communication benefits are insufficient to cover resource costs. The edge node operates autonomously based on the preset economic scheduling rules and local equilibrium control algorithm, without needing to interact with the cloud, thus reducing communication latency and resource consumption.

[0047] If the primary communication trigger instruction is to trigger communication, it is determined that the communication can significantly improve the overall system performance. The cloud-edge communication process is executed immediately, the edge node uploads key status data, receives optimization instructions generated by the cloud based on global information, such as cross-cluster power allocation and charging / discharging strategy adjustment, and executes them in real time to achieve global optimization.

[0048] Decision-making and execution in high-risk scenarios, when the probability of immediate risk... At this time, the system faces safety hazards such as battery overcharging and over-discharging, and voltage exceeding limits, and the principle of safety first takes absolute dominance. At this time, regardless of whether the primary communication trigger command is triggered or not, the command is ignored directly, non-critical loads are forcibly cut off, conservative control laws are switched, and battery balancing or forced cooling systems are started to curb the expansion of risks as soon as possible; at the same time, high-risk alarm information is sent to the cloud, including abnormal state data, risk probability values ​​and emergency measures already implemented, to provide a basis for subsequent global scheduling and fault diagnosis in the cloud.

[0049] The safety arbitration and enforcement module has covered the management and control of routine operation and high-risk emergency scenarios. However, in the actual operation of energy storage power stations, sudden abnormal conditions such as network interruption and partial equipment failure are inevitable. By dynamically adjusting the operation mode and management logic, the system can achieve safety backup, rapid recovery and special adaptation in abnormal scenarios, and ensure the robustness of the system under all operating conditions.

[0050] The abnormal operating condition adaptive module monitors the cloud-edge communication status and device operating status in real time, and executes differentiated adaptive control strategies for different abnormal scenarios to ensure that the system can maintain safe operation or quickly recover to the optimal state under abnormal operating conditions. The specific implementation process is as follows:

[0051] The abnormal operating condition adaptive module judges network connectivity in real time through the communication link status monitoring unit. When a network interruption is detected, such as no communication response for a continuous preset period or a packet loss rate of 100%, the communication suppression mechanism is immediately triggered. At this time, the communication cost model automatically outputs an extreme value response, blocking all communication trigger requests and avoiding invalid communication attempts. The system seamlessly switches to a safe autonomous mode, and the edge nodes rely entirely on the previously built probability management system and local emergency strategy library to operate independently. The state variables are strictly constrained within the safe set by the preset conservative control law to ensure that no safety accidents such as battery overcharging or over-discharging or voltage exceeding the limit occur during the network interruption.

[0052] When the communication link is detected to be restored, and the communication success rate reaches 100% for several consecutive cycles and the latency returns to the normal range, the module will automatically start collecting the current state of charge of each battery cluster, bus voltage, equipment operating status, probability tube boundary, operating mode threshold, and information value calculation weight in the cloud without manual intervention, and complete the synchronization of status and parameters within a preset short period.

[0053] When a local equipment fault is detected through abnormal sensor feedback or a deviation of the state estimate from the threshold, or when a single battery cluster BMS fails, an inverter module malfunctions, or a sensor fails, the module immediately triggers an emergency mode switching command. Simultaneously, core module parameters are dynamically adjusted, the safety factor of the information value decision module is reduced to encourage critical fault information communication, the risk probability calculation cycle of the probability monitoring module is encrypted, and a special fault condition control strategy is initiated to isolate faulty equipment, adjust the power allocation limit of remaining equipment, and activate backup sensor channels.

[0054] Example 1: This example uses a 100MWh lithium-ion battery energy storage power station as a carrier, and presets a safety risk threshold. Confidence level of probability control The complete technical solution is presented as follows: After startup, a stochastic hybrid automaton model identified based on historical data is loaded, and the state vector is set as follows: The corresponding battery state of charge (SOC) is 50%, DC bus voltage is 1.0 pu, output power is 35℃, and the core uses stochastic differential equations to model the continuous dynamics and random disturbances of the equipment. This formula is the core mathematical model of the module, and the deterministic terms in the formula are... The battery charging and discharging characteristics and the power conversion law of the converter are characterized by second-order linear dynamic equations, and the control input is used. This is the current charge / discharge power command; due to stable operating conditions, the disturbance intensity matrix... Fixed values ​​identified from historical data during the pilgrimage process; Wiener process Specifically designed to handle small fluctuations in photovoltaic power (≤3%). After the sensor collects raw data according to the control cycle, the data undergoes preprocessing via a moving average filter, and is then input into an extended Kalman filter algorithm to output a high-precision state estimate. After the residual verification is passed, the estimated value is synchronized to all subsequent modules to provide reliable data support for decision-making and monitoring.

[0055] Local Information Set The system integrates state estimates from nearly 10 control cycles. The core objective of this system—which includes historical instruction execution feedback and local scheduling strategy parameters—is to quantify communication benefits and determine whether to trigger cloud-edge communication, thereby avoiding unnecessary resource consumption.

[0056] This formula is the core of information value quantification, where the optimal cost function is... The weights are set according to the economic optimization model, with the weight ratio of charging / discharging loss to the penalty for deviation from the optimal range being: ; This is incremental information unique to the cloud, specifically the power grid load forecast data for the next hour. The dynamic threshold is generated by coupling two parts: one is the communication cost calculated based on the current 30% bandwidth utilization rate, and the other is the safety factor set under the economic optimization mode. By comparing information value in real time Based on the dynamic threshold, it is determined that the communication benefit outweighs the resource cost, and finally a primary communication instruction is generated, with the synchronization marker instruction having a normal priority.

[0057] A multi-dimensional safety set is defined according to power plant operation and maintenance standards. These correspond to the safe operating ranges of battery state of charge (SOC), DC bus voltage, and battery temperature, respectively. The core is to construct a probability tube through the probability reachability formula to quantify the probability of the system's safe operation in the future time domain. This formula is the core of probability management, where the feasible region of the control input is defined. Limited to charging and discharging power Backward time domain One control cycle, totaling 250ms, with a probability confidence level of [missing information]. The current state estimate Substituting into the formula, it is found that the system state will remain in the safe set for the next 250ms. The lower bound of the probability within is 97%, corresponding to the immediate risk probability. Below the preset safety risk threshold Ultimately, the system outputs a security risk assessment result, providing a rigid basis for security arbitration.

[0058] Combined with the output of the preceding modules, the instantaneous risk probability is calculated. Furthermore, the initial command triggers communication, confirming the execution of cloud-edge collaborative operations. The edge node will then transmit its state estimate. Local operational status data is uploaded to the cloud, and the cloud combines incremental information. (Power grid load forecasting), based on the optimal cost function The system solves for the global optimum and generates a global command to reduce the charging and discharging power to 8MW. After receiving the command, the edge node quickly parses it into a converter modulation signal and executes it. At the same time, it tracks the state changes in real time through an extended Kalman filter algorithm to ensure that the battery state of charge (SOC) is stably maintained in the optimal operating range of 50%, ultimately achieving economically optimized system operation and completing the module closed-loop deployment under this operating condition.

[0059] By achieving precise decision-making through quantitative control, the problem of resource waste in fixed-threshold communication has been solved.

[0060] Example 2: In this example, the core is to achieve risk prediction and emergency triggering through quantitative risk assessment, and fully present the multi-module collaborative emergency management flow.

[0061] Due to cloud cover, the photovoltaic power generation system experienced a sudden power output fluctuation of 12%, which was quickly detected by the terminal sensors. The system then triggered its built-in stochastic hybrid automaton model, initiating a parameter adaptive adjustment mechanism. To address the intensity of the disturbance, the model automatically and synchronously amplified the relevant parameter coefficients representing the stochastic disturbance by a factor of 1.5, effectively enhancing the model's sensitivity to sudden fluctuation signals and ensuring that the dynamic modeling process more accurately matches the dynamic changes in actual operating conditions.

[0062] Meanwhile, the extended Kalman filter algorithm automatically increases its gain coefficient, significantly accelerating the tracking and estimation of the system state. Within just one control cycle, or 50 milliseconds, the algorithm rapidly outputs a high-precision state estimate. According to the estimation results, the current state of charge (SOC) is 79.2%, approaching the safe upper limit of 80%; parameters such as the DC bus voltage and battery temperature remain within normal fluctuation ranges, and no abnormal changes were detected.

[0063] The mode management submodule monitors the above state estimation results in real time. Once it detects that the state of charge (SOC) has entered the safety warning range, it immediately triggers the system operation mode to switch from normal mode to emergency operation mode, and simultaneously sends parameter adjustment instructions to relevant modules such as information value decision-making and probabilistic safety monitoring to adapt to the system's management needs for safety priority strategies in the current state.

[0064] Based on the updated dynamic model, the adjusted parameters of each module, and the current state estimate, the module re-performs a probabilistic safety assessment. The core of this assessment is to quantitatively calculate the probability that the system state will remain within the safe set in the future time domain, constructing a dynamic probability tube to constrain the state trajectory. During the assessment, the feasible domain of the control input remains limited to the charging and discharging power. The system maintains a 5-cycle (250ms) backoff timeout, with the probability confidence level remaining unchanged at 95%. Based on the current SOC estimate of 79.2%, the probability of the SOC exceeding the safety set limit (80%) within the next 250ms is 2.3%, reaching and exceeding the preset 2% safety risk threshold, classifying it as a high-risk condition. The system immediately activates the probability control contraction mechanism, expanding the SOC safety warning interval from the original... Tighten to This strengthens the constraints on state trajectories; at the same time, the frequency of encrypted risk calculations has been adjusted from the usual 50ms / time to 25ms / time, continuously outputting high-risk assessment results in real time, providing continuous and accurate risk support for the security arbitration module.

[0065] Upon receiving a high-risk assessment result from the system, the arbitration module immediately activates the highest-priority emergency response mechanism, strictly adhering to the preset safety priority principle. During this process, the arbitration module completely bypasses the primary communication trigger commands generated by the information value decision module, autonomously executing the local emergency control process. Edge nodes respond rapidly, invoking pre-stored adaptive emergency plans in the local emergency strategy library, generating and issuing control commands to uniformly reduce the current charging and discharging power by 20%. This command achieves a millisecond-level response through the converter, effectively suppressing the continuous rise of the battery's state of charge (SOC), thereby significantly reducing the risk of overcharging at its source.

[0066] Simultaneously, the system sends high-risk alarm information to the cloud monitoring platform. The alarm content fully includes all current key status estimates, real-time calculated risk probability values, and key data such as implemented power control measures, enabling the cloud to comprehensively and in real-time grasp the actual operating conditions and emergency response progress at the edge. After two consecutive control cycles (a total of 100ms) of precise adjustment, the SOC estimate output in real-time based on the extended Kalman filter algorithm decreased to 77.8%, falling back into the tightened safety warning range. At this point, the system detected that the real-time risk probability was below the preset risk threshold and automatically lifted the high-risk warning status.

[0067] To ensure a smooth system transition, the arbitration module continues to operate in emergency mode to cope with potential fluctuations in operating conditions, ensuring that the entire control process maintains high reliability and timely and effective safety management capabilities even under disturbed environments. Through risk assessment, precise decision-making and dynamic adjustment are achieved, significantly improving the system's self-adaptability and safety assurance level under abnormal operating conditions.

[0068] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.

Claims

1. A data-driven collaborative operation and management system for new energy storage power stations, characterized in that, include: The system modeling and state awareness module uses a stochastic hybrid automaton to model the integrated cloud-edge collaborative system of the energy storage power station. It characterizes the continuous dynamics and random disturbances of equipment physical evolution and constructs a stochastic transition model for discrete events in cloud-edge interaction. Combined with the system state definition of multiple operating modes and switching conditions, it acquires real-time state and completes state estimation. The information value decision and primary triggering module calculates the information value of current cloud-edge communication online based on the state estimation results and a local information set containing historical control commands, equipment operating history, and recent disturbance characteristic data. It constructs a dynamic threshold, which is formed by coupling the communication cost corresponding to the current network condition with a safety coefficient adapted to the current operating mode. The information value is compared with the dynamic threshold to generate a primary communication command that will trigger or not trigger. The probabilistic safety monitoring and risk assessment module defines a safety set for the system's critical safety states. Based on a stochastic system model, it calculates the lower bound of the probability that the system state will remain in the safety set within a preset time domain in the future, constructs a probabilistically guaranteed safety evolution boundary, and calculates the instantaneous risk probability of the current state trajectory violating this boundary online in real time. The security arbitration and execution module is used to make final decisions and execute corresponding control strategies based on real-time risk probabilities and primary communication commands; when the abnormal operating condition adaptive module detects a network interruption, it suppresses communication triggering and switches to the edge autonomous mode that depends on the probability-guaranteed security evolution boundary; when the network is restored or a device abnormality is detected, it dynamically adjusts the operating mode and parameters, quickly synchronizes the status and restores collaborative operation or adapts to abnormal operating condition management.

2. The data-driven collaborative operation and management system for new energy storage power stations according to claim 1, characterized in that: The system modeling and state awareness module describes the continuous dynamics and random disturbances of the physical evolution of the equipment through stochastic differential equations, and constructs a stochastic transfer model of discrete events in cloud-edge interaction using Markov decision processes. The multiple operating modes include normal mode, optimization mode and emergency mode, and the switching conditions are determined based on stochastic stability theory and the correlation between system state boundaries.

3. The data-driven collaborative operation and management system for new energy storage power stations according to claim 1, characterized in that: The online calculation of the information value is based on solving a stochastic optimal control problem, and its value is equal to the difference between the expected optimal cost of the system in the scenario of obtaining incremental information through triggered communication and the expected optimal cost of the system in the scenario of relying only on local information; the safety coefficient in the dynamic threshold is dynamically adjusted with the operating mode, and its value in the emergency mode is lower than its value in the normal mode and the optimization mode.

4. The data-driven collaborative operation and management system for new energy storage power stations according to claim 1, characterized in that: The probabilistically guaranteed safety evolution boundary is constructed using a backtracking time-domain stochastic control framework. The safety set includes a voltage safety domain, a state-of-charge safety window, and a temperature threshold range. The instantaneous risk probability is calculated online using an importance sampling fast approximation algorithm.

5. The data-driven collaborative operation and management system for new energy storage power stations according to claim 1, characterized in that: The decision logic of the security arbitration and execution module is as follows: when the immediate risk probability is lower than the preset security risk threshold, the edge autonomy or cloud-edge communication and cloud optimization command response is executed according to the primary communication command; when the immediate risk probability reaches or exceeds the preset threshold, the local emergency control strategy is forcibly activated and a high-risk alarm is sent to the cloud, and this decision has the highest priority.

6. The data-driven collaborative operation and management system for new energy storage power stations according to claim 1, characterized in that: In the abnormal operating condition adaptive module, when the network is interrupted, the edge autonomous mode relies on the probability-guaranteed security evolution boundary and the local emergency strategy library to operate; when the equipment is abnormal, the system automatically switches to emergency mode, and synchronously adjusts the safety factor and risk probability calculation cycle to adapt to the special control requirements of fault conditions.

7. The data-driven collaborative operation and management system for new energy storage power stations according to claim 1, characterized in that: The abnormal operating condition adaptive module detects local equipment faults by monitoring the residual sequence between the system state estimate and the sensor measured value; when the statistical characteristics of the residual sequence deviate from its historical normal fluctuation range for a preset period, it determines that the corresponding equipment has a local fault and triggers a system response.

8. The data-driven collaborative operation and management system for new energy storage power stations according to claim 1, characterized in that: The probability tube constructed by the probability security monitoring and risk assessment module adopts an online rolling update mechanism. The update trigger conditions include the deviation between the current state estimate and the boundary of the probability tube exceeding a preset threshold, and a step change in the intensity of random disturbance. The update cycle is consistent with the system control cycle.

9. The data-driven collaborative operation and management system for new energy storage power stations according to claim 1, characterized in that: The system's expected optimal cost function in the information value decision and primary triggering module It is composed of the weighted cost of charging and discharging loss, the cost of deviating from the safety boundary, and the cost of communication resource consumption. The weight of each cost item is dynamically adjusted based on the current operating mode, with the cost of deviating from the safety boundary having the highest weight in emergency mode.

10. The data-driven collaborative operation and management system for new energy storage power stations according to claim 1, characterized in that: The state vector of the system modeling and state perception module can also be supplemented with adaptation parameters according to the type of energy storage power station. For lithium-ion battery energy storage power stations, battery cell consistency parameters are supplemented, and for flow battery energy storage power stations, electrolyte concentration parameters are supplemented, in order to adapt to the physical characteristics of different types of energy storage devices.

Images