Thermal power generating unit ETS intelligent regulation and control integration method

Abstract

The invention relates to the technical field of thermal power plant automatic control, and discloses a thermal power generating unit ETS intelligent regulation and control integration method, which comprises the following steps: modeling the global state of a unit by adopting a holographic state model based on a tensor network, and calculating to obtain a continuous risk potential field based on an ETS protection boundary; a system Hamiltonian including an economic cost item and a safety risk item determined by the risk potential field is constructed, and an optimal control action considering safety and economy is determined by solving the minimum value of the Hamiltonian. Self-evolution of the system is achieved by actively applying perturbation and utilizing response data to optimize the model online, safety margin is quantified and fused into a unified decision framework, collaborative optimization of safety and economy is achieved, and adaptability and robustness of the system to unknown working conditions are improved by means of active detection capacity.

Description

Technical Field

[0001] This invention relates to the field of automation control technology for thermal power plants, specifically to an integrated method for intelligent control of thermal power unit ETS. Background Technology

[0002] As a key component of the power system, the safety, stability, and economy of thermal power generating units are crucial for ensuring reliable power supply from the grid. The automated control system of modern thermal power units typically consists of two core systems: a distributed control system responsible for routine regulation and optimization control, and an emergency tripping system responsible for performing final safety protection in emergency situations.

[0003] The primary goal of the DCS system and the advanced process control strategies deployed on it is to achieve precise closed-loop control of the unit's complex production processes. This ensures that the unit operates within the economically optimal range, minimizing fuel consumption and pollutant emissions, while stably responding to grid load commands. The ETS, on the other hand, prioritizes the safety of equipment and personnel. It incorporates a series of protection logics that monitor key operating parameters. If a parameter exceeds a pre-set, stringent threshold, it immediately executes mandatory protective actions, such as rapidly cutting off fuel and closing the main steam valve, ultimately shutting down the unit.

[0004] However, in current technological practices, DCS and ETS are separate and independent in terms of design philosophy, control objectives, and operating mechanisms. DCS optimization control algorithms typically treat ETS protection settings as a set of static, insurmountable rigid constraint boundaries. While these algorithms can detect whether these boundaries have been touched during economic optimization, they generally lack a continuous understanding of the degree to which the unit's current operating state "approaches" these safety boundaries, meaning they cannot quantify the potential risk margin implied by the current operating point in real time and accurately.

[0005] This separation of function and objective makes it difficult for the control system to make effective coordinated decisions between safety and economy under complex operating conditions, especially when the load changes rapidly or the equipment is approaching its performance limits. On the one hand, in pursuit of higher economic benefits, the control system may unintentionally maintain the unit's operating point in a region close to a certain ETS protection boundary for an extended period. At this time, the system's ability to resist external disturbances is significantly reduced, and any small, unforeseen fluctuation in operating conditions may easily push the parameters beyond the critical point, thereby triggering unplanned shutdowns and seriously impacting the stable operation of the power grid and the economic benefits of the enterprise.

[0006] On the other hand, in order to avoid such risks, in actual operation, the operation personnel or the control strategy is often forced to take a more conservative operation mode, that is, artificially set a larger safety margin, so that the operating point of the unit is far away from all known ETS boundaries. But this inevitably limits the operation flexibility and adjustable range of the unit, sacrifices the potential economic benefits that can be obtained, and makes the optimization ability of the advanced control system unable to be fully utilized.

[0007] Therefore, there is a lack of an integrated control method in the prior art, which can convert the rigid safety boundary of ETS into a continuous and quantifiable risk index that can be understood and utilized by DCS, and which is inherently and dynamically integrated into the economic optimization decision-making process, so that higher level of collaborative optimization between safety reliability and operation economy can be realized under the premise of ensuring absolute safety. SUMMARY

[0008] In view of the deficiencies of the prior art, the present application provides an intelligent ETS control integrated method for a thermal power generating unit, which solves the problem that in the existing control system of the thermal power generating unit, the rigid safety protection of the emergency trip system and the economic optimization target of the distributed control system are independent of each other, resulting in a lack of collaborative decision-making ability of the system and difficulty in achieving dynamic balance between safety and economy.

[0009] To achieve the above purpose, the present application is realized by the following technical solutions:

[0010] The present application provides an intelligent ETS control integrated method for a thermal power generating unit, which comprises the following steps:

[0011] S1: obtaining real-time operating state variables of the thermal power generating unit;

[0012] S2: based on the real-time operating state variables, using a pre-constructed holographic state model based on a tensor network to model the current global state of the thermal power generating unit;

[0013] S3: based on the current global state and a preset ETS protection boundary of the emergency trip system, calculating a risk potential field of the current state in a multi-dimensional state space;

[0014] S4: constructing a system Hamiltonian including an economic cost term and a safety risk term, wherein the safety risk term is determined by the risk potential field;

[0015] S5: solving an optimal control action that minimizes the system Hamiltonian, and issuing it to the control system of the thermal power generating unit for execution.

[0016] In some embodiments, the tensor network-based holographic state model represents the global state |Ψ(S)> of the thermal power unit in the form of a matrix product state:

[0017]

[0018] Where S is the current state vector of the unit, s i For the discretized value of the i-th state variable, A [i ](s i Let be a third-order tensor associated with the i-th state variable.

[0019] In some embodiments, the calculation of the risk potential field of the current state in the multidimensional state space is specifically implemented by calculating the risk potential energy value Φ(S):

[0020]

[0021] Where S is the current global state point, and B i Let d(S,B) be the hypersurface formed by the i-th ETS protection boundary in the multidimensional state space. i () represents the distance from state point S to boundary surface B i The shortest distance, w i Let α be the risk weight coefficient of the i-th protection boundary, and let α be the potential energy index of the i-th protection boundary.

[0022] In some embodiments, the system Hamiltonian H(U) is in the form of:

[0023] H(U)=H econ (U)+λ·H safe (U);

[0024] Where U is the control action vector to be solved, and H econ 9U) represents the economic cost item, H safe (U) represents the safety risk item, the value of which is determined by the risk potential energy value corresponding to the risk potential field. λ is a dynamic equilibrium factor whose magnitude is positively correlated with the risk potential energy value.

[0025] In some embodiments, finding the optimal control action that minimizes the system Hamiltonian is specifically achieved by employing a simulated quantum annealing algorithm to perform a global optimization within a preset control action space to determine the optimal control action U. * .

[0026] In some embodiments, after solving for the optimal control action, the method further includes a step of activating active detection of safety state perturbations. This step is triggered when: the current risk potential energy value is determined to be lower than a preset safety detection threshold, and the cognitive entropy during the solution process is higher than a preset cognitive entropy threshold due to the existence of multiple alternative control strategies with similar energy.

[0027] In some embodiments, the cognitive entropy E is calculated as follows: First, based on the system Hamiltonian H(U)... j Based on the inverse temperature parameter β, multiple alternative control strategies {U} are calculated. j Boltzmann probability distribution P(U) j ):

[0028] Where Z = ∑ k exp(-βH(U k ));

[0029] Then, based on the Boltzmann probability distribution, its Shannon entropy is calculated to obtain the cognitive entropy E:

[0030] E=-∑ j P(U j )logP(U j );

[0031] In some embodiments, activating active detection of safety state perturbations specifically involves: generating a perturbation control vector δU based on the plurality of alternative control strategies with similar energy; and setting the optimal control action U... * The final control action U is formed by superimposing the perturbation control vector δU on the vector. final The command is then sent to the control system of the thermal power unit for execution.

[0032] In some embodiments, after executing the final control action, the method further includes: acquiring unit state response data caused by the perturbation control vector; and using the state response data to perform online optimization of the model parameters of the holographic state model based on tensor networks.

[0033] Compared with existing technologies, the technical solution provided by this invention has the following beneficial effects: by constructing a holographic state model and a risk potential field, the discrete and rigid safety boundary is transformed into a continuous and computable risk quantification index; by constructing and solving a unified Hamiltonian that includes safety and economic objectives, collaborative optimization decision-making for safety and economic objectives under complex operating conditions is realized; by introducing an active detection and online optimization mechanism, the model can explore and self-improve for unknown or cognitively ambiguous operating conditions, thereby improving the adaptability and decision-making performance of the control system while ensuring unit safety.

[0034] A second aspect of the present invention provides an integrated ETS intelligent control system for thermal power units, the system comprising:

[0035] The status acquisition module is used to acquire the real-time operating status variables of the thermal power unit.

[0036] The holographic state modeling module is configured to model the current global state of the thermal power unit based on the real-time operating state variables and a pre-built holographic state model based on tensor networks.

[0037] The risk potential field calculation module is configured to calculate the risk potential field of the current state in the multidimensional state space based on the current global state and the preset emergency trip system (ETS) protection boundary.

[0038] The control decision module is configured to construct a system Hamiltonian that includes an economic cost term and a safety risk term, wherein the safety risk term is determined by the risk potential field, and the optimal control action that minimizes the system Hamiltonian is obtained by solving the solution.

[0039] The control execution module is configured to send the optimal control action to the control system of the thermal power unit for execution.

[0040] This invention provides an integrated method for intelligent control of ETS (Electronic Toll Collection) in thermal power units. It has the following beneficial effects:

[0041] 1. This invention employs a holographic state model based on tensor networks to perform integrated modeling of the unit's global state. Based on this, a continuous and quantifiable risk potential field is constructed, transforming the discrete and rigid protection thresholds of traditional emergency tripping systems into a continuous safety margin index that can be evaluated throughout the entire operating state space. This enables the system to make forward-looking and global accurate predictions of operational risks, rather than passively responding only after boundaries are reached.

[0042] 2. This invention fundamentally solves the technical problem of the separation between safety and economic objectives in traditional control strategies by constructing a unified system Hamiltonian that includes economic cost and safety risk terms, and introducing a dynamic balance factor linked to the current risk level. This method can adaptively and collaboratively make decisions based on real-time operating conditions within a single optimization framework, finding the economically optimal operating path while ensuring safety margins, thereby improving the control precision and operational flexibility of the unit under complex critical conditions.

[0043] 3. This invention endows the system with self-awareness and online evolution capabilities by introducing cognitive entropy determination and a proactive safety state perturbation detection mechanism. The system can proactively identify unknown or cognitively ambiguous areas of its model and, under the premise of ensuring absolute safety, conduct perturbation exploration to obtain high-value data, thereby optimizing the holographic state model online. This closed-loop learning capability enables the system to continuously improve its cognitive depth of unit characteristics, enhancing its robustness and adaptability when facing new operating conditions it has never encountered before. Attached Figure Description

[0044] Figure 1 This is a flowchart of the method of the present invention;

[0045] Figure 2 This is a system architecture diagram of the present invention. Detailed Implementation

[0046] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0047] Example:

[0048] Please see the appendix Figure 1 This invention provides an integrated method for intelligent control of thermal power unit ETS, comprising the following steps:

[0049] S1: Obtain the real-time operating status variables of the thermal power unit;

[0050] In this embodiment, the steps for obtaining the real-time operating state variables of the thermal power unit are described in detail. This step, as the starting point of the method proposed in this invention, aims to comprehensively and in real-time capture multi-dimensional data that can characterize the overall operating characteristics of the unit, providing a solid data foundation for the subsequent construction of a holographic state model that reflects the deep coupling relationship between various subsystems.

[0051] Specifically, real-time operating status variables are acquired through real-time communication with the distributed control system, monitoring information system, and other related subsystems of the thermal power unit. In a preferred embodiment, the data acquisition cycle should be matched with the dynamic response characteristics of the unit's key parameters to ensure the timeliness and fidelity of the data.

[0052] To ensure the completeness of the state description, a set of N variables was selected, comprehensively covering the key physical processes of the unit. These variables can be divided into several categories, including but not limited to: thermodynamic process variables, such as: main steam pressure, main steam temperature, reheat steam temperature, boiler drum pressure, outlet temperatures of each stage of superheater and reheater, feedwater flow rate and pressure, etc.; mechanical system variables, such as: turbine speed, high / medium / low pressure cylinder exhaust pressure, opening commands and feedback of each major pump valve, shaft vibration, axial displacement, etc.; electrical system variables, such as: generator active power, reactive power, stator voltage, stator current, excitation current, etc.; chemical operating condition variables, such as: boiler feedwater dissolved oxygen, boiler water pH value, steam quality related indicators, etc.

[0053] All N acquired real-time running state variables are organized into a high-dimensional instantaneous state vector S(t) at the same time t, with the following form:

[0054] S(t) = (s1(t), s2(t), ..., s N (t));

[0055] Before inputting the instantaneous state vector S(t) into the subsequent holographic state modeling module, its components need to be preprocessed to eliminate the influence of differences in dimensions and numerical ranges between different physical quantities, ensuring that each variable contributes equally to the model. In one specific implementation, the max-min normalization method is used to normalize each variable s... i The value of (t) is mapped to a uniform interval of [-1, 1].

[0056] Furthermore, to adapt to the subsequent tensor network-based quantized state representation method, the normalized continuous variables need to be discretized. This step transforms the continuous state space into a discrete, finite set of states, which is a necessary preparatory step for constructing a matrix-product state model. For example, the range of values ​​for each normalized variable can be divided into D discrete levels, thereby transforming the original continuous state vector S(t) into a discrete state sequence.

[0057] After the above preprocessing, the resulting discretized state vector can be used as the direct input to a holographic state model based on tensor networks. Each component s of this vector... i Let A be a local tensor corresponding to the matrix product state expression |Ψ(S)>. [i] (s i This initiates the modeling process for the overall state of the unit.

[0058] Through the above methods, this invention not only acquires the apparent operating data of the unit, but also generates structured data input with rich information entropy that can be effectively utilized by advanced models through systematic selection and structured processing, laying the foundation for subsequent accurate risk quantification and intelligent decision-making.

[0059] S2: Based on real-time operating state variables, a pre-built holographic state model based on tensor networks is used to model the current global state of the thermal power unit;

[0060] This embodiment elaborates on the steps of modeling the current global state of a thermal power unit using a pre-constructed holographic state model based on tensor networks, based on real-time operating state variables. This step follows the aforementioned state variable acquisition and preprocessing stages. Its core purpose is to upgrade the processed high-dimensional discrete state vector and transform it into a structured mathematical object that can profoundly represent the complex and nonlinear coupling relationships between various physical quantities within the unit, rather than a simple list of data. This model forms the mathematical foundation for subsequent precise risk quantification and intelligent decision-making.

[0061] In complex industrial systems like thermal power units, numerous state variables exhibit explicit or implicit correlations. For instance, changes in boiler combustion conditions ultimately affect the turbine's vibration state through the steam-water system. Traditional modeling methods struggle to effectively capture these cross-system, long-distance correlations. To address this, this invention introduces a tensor network method derived from quantum many-body physics to construct a holographic state model. This model efficiently represents a high-dimensional state space and encodes the correlation structure between variables within a compact mathematical framework.

[0062] In a preferred embodiment, the holographic state model based on tensor networks takes the form of a matrix product state. For the N-dimensional discrete state vector S = (s1, s2, ..., s...) obtained in the previous step... N The global quantized state |Ψ(S)> corresponding to it in the model is represented as a matrix product of a series of local tensors:

[0063]

[0064] Here, the physical meaning of each symbol in the formula is explained: |Ψ(S)> represents a holistic, structured mathematical description of the unit under a specific global state S, i.e., a holographic state. i A represents the discretized value of the i-th state variable. [i] (s i ) is related to the i-th state variable and its value s iThe core component is essentially a third-order tensor. These two virtual indices act like a chain, connecting the tensors of the i-th variable with those of the (i-1)-th and (i+1)-th variables. Through this series of multiplication operations, the correlation between local variables is passed down level by level, ultimately forming a unified state representation that can characterize the global correlation properties of the entire system.

[0065] A key design parameter in the model is the bond dimension χ. This parameter defines the dimension of the virtual indexes connecting adjacent local tensors. Physically, the size of the bond dimension χ determines the complexity and distance of the relationships between variables that the model can capture and represent. A larger χ value enables the model to describe deep coupling relationships between two variables that are far apart in the system, thus more closely approximating the complex characteristics of real industrial processes. The value of this parameter is selected during the model building phase based on system complexity and computational resources.

[0066] The holographic state model is pre-built, which means that an offline training phase is needed to determine all local tensors A in the model before the system is put into online operation. [i] The specific numerical value. In one specific embodiment, the training process utilizes a large amount of historical operating data, especially those data records marked as safe and stable operating conditions. Variational algorithms such as the density matrix renormalization group can be used to iteratively optimize and solve for a set of optimal tensors A. [i] This allows the constructed model to best reproduce and represent the set of normal operating states.

[0067] During the real-time operation of the system, whenever a new state vector S(t) is acquired, the model performs a forward computation. That is, according to the above formula, the components of S(t) are substituted, and tensor multiplication is used to quickly obtain the global state representation |Ψ(S(t))> at the current moment. This output |Ψ(S(t))>, as a mathematical object rich in structural information, will be passed to the subsequent risk potential field calculation module for a more accurate and profound assessment of the system's security status.

[0068] S3: Based on the current global state and the preset emergency trip system (ETS) protection boundary, calculate the risk potential field of the current state in the multidimensional state space;

[0069] In this embodiment, the steps for calculating the risk potential field of the current state in the multidimensional state space based on the current global state and the preset emergency trip system (ETS) protection boundary are described in detail. This step is a key link connecting system state perception and intelligent decision-making. Its core task is to transform the high-dimensional, structured holographic state model output from the previous step into an intuitive, quantifiable scalar risk indicator that can guide control decisions.

[0070] Traditional safety protection mechanisms assess risk in a discrete and binary manner: the state is either within the safe zone or has reached a boundary, triggering protection. This approach fails to describe the degree and rate at which a state approaches a dangerous boundary. Therefore, this invention introduces the concept of a probabilistic safety manifold, mathematically represented as a continuous scalar potential field defined over the entire N-dimensional state space. This risk potential field serves to assign a definite risk potential value to each possible global state point of the unit, thereby concretizing the abstract concept of safety margin.

[0071] Specifically, this step first performs mathematical geometrization on all inherent ETS protection boundaries in the thermal power unit, which serve as the ultimate safety baseline, in an N-dimensional state space. For example, a simple single-variable protection setting, such as the upper limit of main steam pressure, is mapped as a hypersurface in N-dimensional space. All i ETS protection boundaries together constitute a set of boundary surfaces {B} representing a set of prohibited regions in the state space. i}

[0072] Having obtained the current global state point S and the aforementioned boundary surface set {B}, i Afterwards, the system calculates the risk potential energy Φ(S) of the current state in real time. This calculation integrates the distances from the current state point to all different protection boundaries and weights them according to the importance of each boundary. In a preferred embodiment, the formula for calculating the risk potential energy Φ(S) is as follows:

[0073]

[0074] Here, the various components of the formula and their technical implications are explained: d(S,B i ) represents the distance from the current state point S to the i-th boundary surface B. i The shortest distance. This distance is a direct measure of safety margin, and its calculation method can be flexibly selected according to the geometric characteristics of the boundary. For example, for linear boundaries, the distance formula from a point to a hyperplane can be used, while for nonlinear boundaries, a constrained optimization problem is solved. The calculation of this distance value unifies the multidimensional state of the unit under the common metric of safety distance.

[0075] This is the core functional form of this method. It employs the inverse power of the distance to construct a potential field with repulsive properties. When the state point S is far from the boundary B... i When the distance d is large, the potential energy contributed by this term approaches zero; however, when the state point S approaches the boundary B infinitely... i When the distance d approaches zero, the potential energy contributed by this term increases sharply, approaching infinity. This nonlinear, steep increase can effectively simulate the drastic changes in risk near a safety boundary in the real world.

[0076] α i Let be the potential energy exponent of the i-th protective boundary, and be a constant greater than 0. This parameter is used to adjust the risk potential field at boundary B. i The steepness of the surrounding terrain. For critical protection systems with high dynamic response requirements and stringent margin requirements, a larger α value can be set. i The value causes the risk potential energy to increase at a higher order as the distance decreases, thus reflecting a more respectful control tendency towards the boundary in decision-making.

[0077] w i This represents the risk weighting coefficient for the i-th protection boundary. The setting of this coefficient reflects the differences in safety level and economic impact among different protection actions. For example, protection actions that may damage main equipment should be assigned a much higher weight than those for general auxiliary equipment protection. These weighting coefficients can be determined based on failure mode and effects analysis or long-term expert operating experience, ensuring that the final risk potential value comprehensively reflects multiple risk considerations.

[0078] Finally, by summing Σ over all the potential energy terms generated by the protective boundaries... i The total risk potential energy value Φ(S) is obtained. This scalar value does not reflect the degree of danger of a single parameter in isolation, but rather is a holistic and global quantitative assessment of the combined effects of all potential risk sources under the current overall state of the unit.

[0079] The risk potential energy Φ(S), as the final output of this step, will be passed to the subsequent quantum-inspired decision-making module. In this module, it will be directly used as the security risk term H in the system's Hamiltonian. safe This becomes the inherent force driving the control system to avoid risks and seek safety, thus achieving a smooth connection from state perception to decision-driven action.

[0080] S4: Construct a system Hamiltonian that includes economic cost terms and security risk terms, where the security risk terms are determined by the risk potential field;

[0081] In this embodiment, the steps for constructing the system Hamiltonian, which includes economic cost and safety risk terms, are described in detail, with the safety risk term determined by the risk potential field. This step is the foundation of the core decision-making process of this invention. Its purpose is to unify the safety risk quantified in the preceding steps with the economic objective of unit operation into a self-consistent and optimizable mathematical framework. This transforms the two conflicting and hierarchical decision-making objectives in traditional control into a unified single optimization problem seeking the global optimal solution.

[0082] To achieve this objective, this invention draws upon the concept of the Hamiltonian in theoretical physics, extending it to a target function describing the generalized total cost or total energy of a system. The constructed system Hamiltonian H(U) is a function of the future control action vector U to be solved, and its form is a weighted sum of economic cost and safety risk terms:

[0083] H(U)=H econ (U)+λ·H safe (U);

[0084] Here, we will provide an in-depth explanation of each component of the Hamiltonian formula and its technical implications.

[0085] First, security risk item H safe (U), whose value is directly and uniquely determined by the risk potential field calculated in the previous stage. Specifically, it is defined as the future state S that the unit will reach after executing a hypothetical control action vector U. ′ The corresponding risk potential value:

[0086] H safe (U)=Φ(S ′ );

[0087] The technical significance of this design lies in transforming the aforementioned potential field describing the static risk distribution into a direct constraint on future dynamic behavior. Any control action U that might cause the unit's state to migrate to a high-risk region will inevitably lead to a higher H. safe (U) value. Therefore, this security risk term plays the role of a repulsive barrier in the Hamiltonian. In the subsequent optimization process, the algorithm will naturally avoid control strategies that would increase this term, thus inherently achieving the goal of risk avoidance.

[0088] Secondly, the economic cost item H econ (U) serves to quantify the economic advantages and disadvantages of the unit's operating state. In a preferred embodiment, this term is defined as the future state S after executing control action U. ′ Its economic indicators are consistent with the preset economic optimal state S. opt The degree of deviation between them. Its mathematical form can be expressed as:

[0089]

[0090] Among them, S opt It is a predetermined ideal state vector that represents the best economic performance under the current load. It can be obtained from the unit's design basis, performance test data, or offline optimization calculations. H is a cost mapping function that maps a high-dimensional state vector S to one or more scalar economic indicators, such as power generation coal consumption rate, plant power consumption rate, or pollutant emission concentration. Therefore, H econ (U is essentially the potential energy of the system in the economic dimension. The lower the value, the closer the operating state is to the economic optimal goal.)

[0091] Furthermore, the dynamic balance factor λ is the core mechanism for achieving coordinated decision-making between safety and economy. Unlike a fixed weighting coefficient, the value of λ changes dynamically and is strongly correlated with the current risk level of the unit, i.e., the risk potential energy Φ(S) of the current state S. In a specific embodiment, λ is designed as a monotonically increasing function of Φ(S).

[0092] The technical essence of this dynamic adjustment mechanism lies in the fact that when the unit operates in a deep safety zone far from all safety boundaries, the current risk potential energy Φ(S) is extremely low, thus making the value of λ correspondingly smaller. At this time, the magnitude of the system Hamiltonian H(U) is mainly determined by the economic cost term H... econ (U) determines that the main task of the optimization algorithm will be to find the control action that maximizes economic efficiency. Conversely, as the unit's operating state gradually approaches any safety boundary, the current risk potential energy Φ(S) increases significantly, and the value of λ also increases sharply. At this time, the safety risk term λ·H safe (U) will dominate the Hamiltonian. The optimization algorithm will automatically and smoothly switch to prioritizing the search for control actions that can reduce risk potential energy the fastest, even if such actions sacrifice some economic performance in the short term.

[0093] The Hamiltonian of the system constructed in the above manner is no longer a simple combination of two independent objectives, safety and economy, but an organic unity that can adaptively adjust the decision focus according to the real-time risk level. It rigorously transforms the complex, multi-objective control decision problem into a well-defined physical optimization problem of finding the minimum value of a single scalar function, paving the way for subsequent solutions using global optimization algorithms to find the optimal control strategy.

[0094] S5: Solve for the optimal control action that minimizes the Hamiltonian of the system and send it to the control system of the thermal power unit for execution.

[0095] In this embodiment, the steps of finding the optimal control action that minimizes the system Hamiltonian and issuing it to the control system of the thermal power unit for execution are described in detail. This step is the decision-making center and execution exit of the method proposed in this invention. It inherits the unified system Hamiltonian constructed in the previous steps, which integrates safety and economic objectives, and aims to calculate a specific and executable sequence of control commands that can achieve the best balance under the current operating conditions through efficient global optimization.

[0096] Specifically, the core of this step is an optimization process. In the previous step, the system Hamiltonian H(U) has been defined as a scalar function of the future control action vector U. Therefore, the problem of finding the optimal control action is rigorously transformed into a mathematical global optimization problem:

[0097]

[0098] Among them, U * This is the optimal control action vector that needs to be solved. It should be noted that due to the highly nonlinear and strongly coupled characteristics of thermal power unit systems, as well as the non-convexity of the risk potential field function, the functional topology of the constructed system Hamiltonian H(U) is usually extremely complex, with a large number of local minima. Traditional gradient-based optimization algorithms are prone to getting trapped in these local optima and cannot guarantee the global optimality of the decision.

[0099] Therefore, in a preferred embodiment, a simulated quantum annealing algorithm is used to perform the above solution process. This algorithm draws on the physical process of quantum annealing, introducing the simulation of the quantum tunneling effect, enabling it to overcome high potential barriers and effectively escape local minima during the optimization process. This gives the SQA algorithm a significant advantage over traditional optimization methods when dealing with such complex, non-convex global optimization problems, and it is more likely to find the solution U. * It is either the globally optimal solution or a high-quality suboptimal solution.

[0100] Optimal control action U * Specifically, it is a vector containing the adjustment amounts of multiple control variables over one or more future time steps.

[0101] Furthermore, in determining the optimal control action U * Subsequently, this method does not directly issue the control action in all situations. Instead, it enters a pre-judgment stage to determine whether to activate the active detection mechanism for safety state perturbations. As before, the system determines whether the current risk potential and cognitive entropy simultaneously meet preset triggering conditions. If the triggering conditions are not met, the final control action U is then issued. final That is, the optimal control action U obtained through optimization. * If the triggering condition is met, the system will generate an additional perturbation control vector δU and superimpose it with the optimal solution to form the final control action U. final =U * +δU. This measure aims to ensure that the system possesses the ability to proactively explore and learn on its own, while maintaining absolute safety.

[0102] Finally, the final control action U finalThis will be sent to the control system of the thermal power unit for execution. In a specific embodiment, this sending process manifests as the vector U... final The control adjustment quantities represented by each component are transformed into the increments or new target values ​​of the corresponding basic control loops in the distributed control system of thermal power units. These new setpoints are then precisely and stably tracked and executed by the basic loop controllers of the DCS, thereby translating the top-level intelligent decisions into the physical actions of the bottom-level actuators.

[0103] It is important to emphasize that the entire decision-making and execution process described in this invention remains under the existing, physical emergency tripping system. The Emergency Response System (ETS), as the unit's final and insurmountable safety barrier, operates independently and with the highest priority. Any control commands output by this invention are only effective within the protection boundaries of the ETS, aiming to ensure that the unit's operating state remains far from the hard threshold that triggers ETS action through predictive intelligent control, rather than replacing or interfering with the ETS's ultimate protection function.

[0104] Please see the appendix Figure 2 The integrated ETS intelligent control system for thermal power units includes the following steps:

[0105] The status acquisition module is used to acquire the real-time operating status variables of the thermal power unit;

[0106] The holographic state modeling module is configured to model the current global state of the thermal power unit based on a pre-built holographic state model based on tensor networks, using real-time operating state variables.

[0107] The risk potential field calculation module is configured to calculate the risk potential field of the current state in the multi-dimensional state space based on the current global state and the preset emergency trip system (ETS) protection boundary.

[0108] The control decision module is configured to construct a system Hamiltonian that includes economic cost terms and safety risk terms. The safety risk terms are determined by the risk potential field, and the optimal control action that minimizes the system Hamiltonian is obtained by solving the problem.

[0109] The control execution module is configured to send the optimal control action to the control system of the thermal power unit for execution.

[0110] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An integrated method for intelligent control of thermal power unit ETS, characterized in that, Includes the following steps: S1: Obtain the real-time operating status variables of the thermal power unit; S2: Based on the real-time operating state variables, a pre-constructed holographic state model based on tensor networks is used to model the current global state of the thermal power unit; S3: Based on the current global state and the preset emergency trip system (ETS) protection boundary, calculate the risk potential field of the current state in the multidimensional state space; S4: Construct a system Hamiltonian that includes an economic cost term and a security risk term, wherein the security risk term is determined by the risk potential field; S5: Solve for the optimal control action that minimizes the Hamiltonian of the system and send it to the control system of the thermal power unit for execution.

2. The method according to claim 1, characterized in that, The tensor network-based holographic state model represents the global state of the thermal power unit in the form of matrix product states.

3. The method according to claim 1, characterized in that, The calculation yields the risk potential field of the current state in the multidimensional state space, specifically as follows: By calculating the distance from the current global state point to each preset ETS protection boundary, and using preset risk weights and potential energy exponents, the risk potential energy value corresponding to the risk potential field is obtained by superimposing the weighted distances according to the reciprocal power function.

4. The method according to claim 1, characterized in that, The economic cost term and the safety risk term in the system Hamiltonian are weighted by a dynamic balance factor, the magnitude of which is positively correlated with the risk potential energy value corresponding to the risk potential field.

5. The method according to claim 1, characterized in that, The optimal control action that minimizes the Hamiltonian of the system is determined by employing a simulated quantum annealing algorithm to perform a global search within a preset control action space.

6. The method according to claim 1, characterized in that, After solving for the optimal control action, the following steps are also included: Determine whether the current risk potential value is lower than the preset safety detection threshold, and whether there are multiple alternative control strategies with similar energy during the solution process that cause the cognitive entropy to be higher than the preset cognitive entropy threshold; If the judgment result is yes, then activate the active detection of safety state perturbations.

7. The method according to claim 6, characterized in that, The cognitive entropy is calculated as follows: Based on the system Hamiltonian and inverse temperature parameters, the Boltzmann probability distributions of multiple alternative control strategies are calculated. Based on the Boltzmann probability distribution, its Shannon entropy is calculated to obtain the cognitive entropy.

8. The method according to claim 7, characterized in that, The activation of active detection of security state perturbations specifically includes: Based on the multiple alternative control strategies with similar energy, a perturbation control vector is generated; The optimal control action is superimposed with the perturbation control vector to form the final control action, which is then sent to the control system of the thermal power unit for execution.

9. The method according to claim 8, characterized in that, After executing the final control action, the following is also included: Collect unit state response data caused by the perturbation control vector; The model parameters of the tensor network-based holographic state model are optimized online using the state response data.

10. An integrated intelligent control system for thermal power unit ETS, comprising the integrated intelligent control method for thermal power unit ETS according to any one of claims 1-9, characterized in that, Includes the following steps: The status acquisition module is used to acquire the real-time operating status variables of the thermal power unit. The holographic state modeling module is configured to model the current global state of the thermal power unit based on the real-time operating state variables and a pre-built holographic state model based on tensor networks. The risk potential field calculation module is configured to calculate the risk potential field of the current state in the multidimensional state space based on the current global state and the preset emergency trip system (ETS) protection boundary. The control decision module is configured to construct a system Hamiltonian that includes an economic cost term and a safety risk term, wherein the safety risk term is determined by the risk potential field, and the optimal control action that minimizes the system Hamiltonian is obtained by solving the solution. The control execution module is configured to send the optimal control action to the control system of the thermal power unit for execution.

Images