Tower type photo-thermal power station heat absorber layered coordinated control system and method

Abstract

This invention belongs to the field of temperature control technology and discloses a hierarchical coordinated control system and method for receivers in tower-type solar thermal power plants. It employs a temperature gradient-driven adaptive partitioned acquisition strategy, initially generating a basic partition with a fine center and coarse edges based on optical simulation. During operation, the partitions are dynamically refined or merged according to local temperature gradients, achieving refined temperature measurement and control at the partition level. By constructing a linear state-space model with online parameter correction and combining intrinsic orthogonal decomposition to reduce the solution scale, the invention maintains constraint verification accuracy while ensuring solution efficiency. It is equipped with dual-dimensional control barrier function constraints for temperature amplitude and rate of change, setting two levels of safety thresholds and dynamically adjusting convergence characteristics according to the safety margin. This effectively mitigates local temperature differences under high irradiance and cloud transient conditions, controlling surface temperature deviation within a narrow range, avoiding local overheating and thermal fatigue damage, and extending the service life of the receiver tubes.

Description

Technical Field

[0001] This invention belongs to the field of temperature control technology, specifically a hierarchical coordinated control system and method for the receiver of a tower solar thermal power plant. Background Technology

[0002] Tower-type concentrated solar power (CSP) is a baseload power source with long-term heat storage capacity in clean energy systems. The receiver, as the core of energy conversion in a CSP plant, directly determines the service life and power plant efficiency through the uniformity of its surface temperature distribution and the stability of the outlet molten salt temperature. Temperature control of tower receivers has long faced three major technical challenges: uneven spatial distribution, large time delays, and strong coupling of multiple variables. Existing temperature control technologies mainly have the following limitations: The surface temperature control scheme based on fixed grouping of heliostats achieves a rough temperature distribution by pre-dividing the mirror field into groups and targeting different areas of the receiver with different groups. Essentially, it is an open-loop proportional control. Its grouping strategy is fixed and cannot adapt to dynamic changes in DNI (Difference of Noise), resulting in a large surface temperature difference in steady state. Under transient cloud conditions, the risk of local overheating is prominent, and it cannot achieve fine-grained temperature regulation at the zone level.

[0003] The traditional PID-based single-loop control scheme for outlet temperature uses the outlet temperature deviation as input and achieves closed-loop control by adjusting the molten salt flow rate. However, molten salt has a large thermal inertia and is a typical controlled object with a large time delay. When the DNI (Discharge Temperature Indicator) changes abruptly, the traditional PID response lags significantly, resulting in a large outlet temperature overshoot. Under extreme conditions, this can easily trigger an over-temperature interlock shutdown.

[0004] In existing power plants, distributed dual-loop independent control schemes typically deploy surface temperature control and outlet temperature control as two independent loops. However, these two loops are strongly coupled: adjusting the heliostat aiming point simultaneously alters the surface heat flux distribution and total energy input, while adjusting the molten salt flow rate affects both the outlet temperature and surface heat transfer conditions. Under dynamic operating conditions, the high degree of coupling between the two loops makes the distributed control architecture prone to problems such as control command conflicts and regulation oscillations.

[0005] Furthermore, existing technologies have not yet established a system-level energy conservation monitoring and target conflict arbitration mechanism. When the targets of the two control loops conflict, there is a lack of unified decision-making logic to achieve a performance trade-off under the premise of safety priority. At the same time, timing prediction technology is mostly used only as an auxiliary monitoring tool and has not formed a complete closed-loop collaborative link with feedforward compensation and feedback control, so the technical advantages of predictive control have not been fully released. Summary of the Invention

[0006] The purpose of this invention is to provide a layered coordinated control system and method for receivers in tower solar thermal power plants, in order to solve one or more of the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a layered coordinated control system for the receiver of a tower solar thermal power plant, including a temperature field zone acquisition module, a temperature field prediction and control module, an irradiation prediction and correction module, an outlet temperature regulation module, a dual-loop decoupling module, a thermal balance anomaly control module, and a multi-objective arbitration module; Furthermore, the temperature field zoning acquisition module uses an infrared thermal imager to form an area array acquisition array to measure the temperature of the entire surface of the receiver. It adopts a temperature gradient-driven adaptive zoning strategy. In the initial state, the heat flux density distribution on the surface of the receiver under different irradiation conditions is pre-calculated through optical simulation to generate a basic zoning scheme. The central high-energy flow area is divided into fine control zones, and the edge low-energy flow area is divided into coarse-grained control zones. The total number of zones is configured according to the power plant capacity. During operation, the temperature gradient values ​​of each region are calculated in real time. When the local temperature gradient exceeds the set threshold, the corresponding region is automatically subdivided. When the local temperature gradient remains below the set threshold, adjacent low-gradient regions are automatically merged. The total number of regions is adjusted within a preset range. Temperature gradient calculation adopts the partition boundary neighborhood difference method. The temperature sampling values ​​of two rows of pixels on both sides of the shared boundary of adjacent partitions are extracted, the difference between the average values ​​of the sampling on both sides is calculated, and the difference is divided by the physical distance corresponding to the boundary to obtain the temperature gradient value of the boundary. The maximum value of the gradient of all adjacent boundaries of a single partition is taken as the local temperature gradient of the partition.

[0008] The collected raw temperature data is processed sequentially through non-uniformity correction, noise filtering, and coordinate mapping. The average temperature, maximum temperature, and temperature gradient of each zone are statistically obtained according to the current effective zone and output to the temperature field prediction and control module.

[0009] Furthermore, the temperature field prediction and control module establishes a linear state-space model of the absorber surface temperature, using the average temperature of each zone as the state vector and the zone heat flow adjustment amount corresponding to each group of heliostats as the control input. The initial state transition matrix and input matrix are obtained by linearizing the energy conservation equation at the rated operating point. During operation, the parameters of the linear state-space model are corrected online using the subspace identification method and the recursive least squares method to compensate for parameter drift. Safety constraints for control barrier functions are constructed, defining two types of safety boundary functions for each partition: temperature amplitude and temperature change rate. Two levels of safety thresholds are set for temperature amplitude: the first threshold is 95% of the limit temperature of the tube screen material, and the second threshold is 92%. Upper limits for single-cycle temperature rise and fall rates are set for temperature change rate. Both types of constraints adopt the form of exponential discrete CBF constraints, and the convergence time constant is adjusted according to the current temperature safety margin. Safety constraint formula for discrete control barrier function: This represents the safety boundary function value of the i-th control zone at time k+1, indicating the margin between the zone temperature and the safety threshold after the control action is executed; a value greater than 0 indicates that the temperature is within the safe range, and a smaller value indicates that it is closer to the over-temperature boundary.

[0010] The value of the safety boundary function for the i-th control zone at time k represents the measured margin of the zone temperature from the safety threshold at the current time, which is calculated from the current zone temperature and the safety threshold.

[0011] This indicates the control zone number, corresponding to each independent temperature control area divided on the surface of the absorber.

[0012] The sequence number represents the control time, indicating the kth control cycle.

[0013] The system state vector at time k is composed of the current average temperature of all control zones on the surface of the receiver and is the input variable of the safety boundary function.

[0014] The system state prediction vector at time k+1 is calculated from the state-space model. It represents the temperature state after the control increment is executed and is the prediction input variable of the safety boundary function.

[0015] This represents the CBF decay coefficient, with a value ranging from 0.01 to 0.10. It is used to constrain the decay rate of the safety boundary function. The larger the coefficient, the stronger the callback constraint when the temperature approaches the safety boundary, which can prevent the temperature from exceeding the safety threshold.

[0016] This represents the sampling period of the control system, which is the time interval between two adjacent control moments. The typical value for a 100 MW tower solar thermal power plant is 500–1000 ms.

[0017] A snapshot matrix is ​​constructed by continuously collecting temperature field data under normal operating conditions. The snapshot matrix is ​​updated at fixed intervals. The first m modes with an energy ratio of more than 95% are selected by singular value decomposition to construct the projection matrix. The intrinsic orthogonal decomposition technique is used to complete the mode order reduction, reduce the dimension of the original state space, and reduce the solution scale. The solution process adopts a collaborative mechanism of order reduction optimization and full-dimensional dual-dimensional verification. After obtaining the control increment by completing the rolling optimization solution of model predictive control in the low-dimensional space, the control increment is mapped back to the original high-dimensional space through the projection matrix. The two types of CBF constraints, namely temperature amplitude and temperature change rate, are verified point by point for all partitions. If a constraint violation occurs, the constraint projection method is used to correct the control increment to the feasible region boundary. Combined with a warm-start strategy and multi-threaded parallel computing, the single-cycle solution time of the partitioned system is shortened.

[0018] Model predictive control rolling optimization objective function: This represents the objective function value of the model predictive control. The optimization objective is to minimize this value in order to achieve a balance between temperature control accuracy and the smoothness of the regulation action.

[0019] This represents the prediction time domain, which is the number of control steps the controller takes to predict temperature changes. The value ranges from 10 to 20. The longer the time domain, the wider the prediction coverage and the more accurate the temperature trend prediction, but the computational load increases accordingly.

[0020] This represents the system state prediction vector at the j-th prediction step, corresponding to the temperature prediction results of all partitions at the j-th future time, used to predict temperature trends and constraint compliance.

[0021] The penalty weight coefficient represents the control increment and is used to balance the accuracy of temperature control with the intensity of heliostat adjustment. The larger the weight, the stronger the constraint on the adjustment, the smoother the control process, and the less frequent and large movements of the heliostat.

[0022] This represents the control time domain, i.e., the number of control increment steps in the optimization solution output, with a value range of 3 to 8. Under the rolling optimization mechanism, only the first control increment is executed, and subsequent step sizes are used to support multi-step iterative optimization.

[0023] This represents the control input increment for the j-th control step, corresponding to the offset adjustment of the aiming point of each heliostat, which directly determines the increase or decrease of the incident heat flux in each zone.

[0024] The hot-start strategy uses the optimal control increment sequence obtained from the previous control cycle as the initial iteration point for the current optimization solution, while inheriting the constraint feasible region determination result from the previous cycle to reduce the number of initial iteration steps. Parallel computing divides all partitions into multiple independent verification groups, each group is assigned to an independent computing thread, and synchronously performs point-by-point verification of the constraints on temperature amplitude and temperature change rate. After each thread completes its verification, it summarizes the verification results and information on partitions that violate the constraints, and performs subsequent correction operations in a unified manner. The number of threads is configured to match the number of hardware computing cores.

[0025] Furthermore, the irradiation prediction and correction module adopts a sequence-to-sequence architecture long short-term memory network to realize multi-step prediction of normal direct irradiance. The network includes two sub-networks: an encoder and a decoder. The encoder inputs the DNI values ​​and exit temperature values ​​from the past 4 to 8 steps to extract temporal features; the decoder outputs a DNI prediction sequence for the next 5 to 15 steps based on the hidden state output by the encoder. A three-level robustness enhancement mechanism is constructed. The first level is the prediction confidence interval estimation. During the inference stage, the Dropout layer is kept active. Predictive samples are generated through Monte Carlo sampling. The prediction mean and 95% confidence interval are calculated. When the interval width exceeds the threshold, the lower bound value is used as the feedforward input. The second stage is error autoregressive correction, which establishes a moving average model of historical prediction errors to estimate system deviations and correct prediction outputs in real time. The third level is the extreme weather degradation strategy. When a step change in DNI is detected to exceed a preset threshold in a short period of time, the prediction output is paused and switched to the measured value feedforward mode. A temperature deviation reverse correction channel is constructed to convert the deviation between the measured value of the outlet temperature and the value calculated by the heat balance model into an equivalent heat flow deviation, and the heat flow conversion factor corresponding to the DNI prediction sequence is corrected in real time.

[0026] Furthermore, the outlet temperature regulation module adopts a three-level architecture of prediction correction, phase feedforward, and fuzzy feedback. The first level is the prediction correction stage, where the irradiation prediction correction module provides the future DNI change trend and heat flow conversion correction coefficient. The second level is the phase feedforward stage, which calculates the average residence time of molten salt in the absorber tube in real time based on the energy conservation equation and the molten salt flow temperature hysteresis model. It aligns the DNI prediction sequence with the molten salt temperature hysteresis phase, determines the lead amount and regulation rate of molten salt flow pre-regulation, and completes the flow pre-regulation synchronously with the heat flow change reaching the outlet section. The molten salt flow feedforward compensation amount is calculated based on the DNI prediction change. Formula for calculating molten salt flow feedforward compensation: This represents the feedforward compensation amount of the molten salt mass flow rate, which is the molten salt flow rate value adjusted in advance according to the irradiation change. It is used to offset the control lag caused by the thermal inertia of the molten salt and to achieve phase alignment between the irradiation change and the flow rate adjustment.

[0027] The effective heat absorption area of ​​the receiver is the actual effective area of ​​the receiver surface that can receive sunlight concentrated by the heliostat, which is determined by the structural parameters of the power plant equipment.

[0028] The heat absorption efficiency of the absorber is represented by a value ranging from 0.80 to 0.90, which indicates the proportion of incident solar radiation energy that is ultimately converted into molten salt heat energy, taking into account comprehensive losses such as optical losses, convective and radiative heat losses.

[0029] The predicted change in direct normal irradiance (DNI) is output by a time-series deep learning network. It represents the fluctuation of solar irradiance intensity over a future period and serves as the input for feedforward adjustment.

[0030] This represents the isobaric specific heat capacity of molten salt, which is the amount of heat required to raise the temperature of a unit mass of molten salt by a unit. It is an inherent thermophysical parameter of the molten salt working fluid.

[0031] This indicates the set inlet and outlet temperature rise of molten salt flowing through the absorber, which is the difference between the outlet temperature and the inlet temperature required by the power plant design. It is a fixed operating set value.

[0032] The third stage is a fuzzy feedback loop, which uses a piecewise adaptive fuzzy proportional-integral-derivative controller. The outlet temperature deviation and the rate of change of deviation are inputs, and the three PID parameters are adjusted online through fuzzy inference. Temperature deviation is divided into three segments: normal, warning, and over-limit. Each segment is configured with an independent fuzzy rule base and parameter adjustment range. In the over-limit segment, the differential action weight is forcibly increased. The fuzzy inference system adopts a dual-input, three-output structure. The input variables, temperature deviation and deviation change rate, are divided into five fuzzy subsets. The output variables, proportional coefficient, integral coefficient, and differential coefficient, are continuously adjustable within the set range. The centroid method is used to complete the defuzzification calculation.

[0033] Furthermore, the dual-loop decoupling module establishes a real-time temperature dynamic relative gain array monitoring mechanism. In each control cycle, based on the gain matrix of the control quantity to temperature under the current operating condition, the temperature RGA matrix is ​​calculated in real time through the relative gain array formula to quantify the coupling degree between the surface temperature uniformity control loop and the outlet temperature stability control loop. When the minimum value of the main diagonal element of RGA is greater than or equal to 0.8, the pairing relationship between the main diagonal controlled variable and the control variable of surface temperature-heat flow regulation and outlet temperature-molten salt flow is maintained, a distributed control mode is adopted, and the decoupling compensation coefficient is set to zero. When the minimum value of the main diagonal element is in the range of 0.35 to 0.8, the continuous decoupling enhancement strategy is activated. The decoupling compensation coefficient is adjusted linearly and continuously with the RGA value. The decoupling compensator matrix is ​​calculated based on the current RGA matrix. The temperature cross-coupling effect between the loops is offset by feedforward compensation. At the same time, joint limiting constraints are applied to the control increments of the two loops. When the minimum value of the main diagonal element is less than 0.35, the system switches to centralized control mode, and the multi-objective arbitration module generates heat flow regulation and molten salt flow instructions in a unified manner. Formula for calculating relative gain array matrix: This represents the relative gain array (RGA) matrix, used to quantify the degree of coupling between each control input and the controlled output. The closer the value of the main diagonal element of the matrix is ​​to 1, the smaller the coupling interference between the corresponding input and output pairs, and the more suitable it is for distributed control.

[0034] This represents the steady-state gain matrix of the system. The value at each position in the matrix represents the steady-state influence of the corresponding control input on the corresponding controlled output. In this dual-loop control system, it fully includes the "influence of heat flow regulation on surface temperature", the "influence of molten salt flow rate on outlet temperature", and the two sets of cross-coupled influence relationships.

[0035] Represents the steady-state gain matrix The inverse matrix is ​​used to derive the interaction coefficients between control loops and quantify the degree of indirect interference of one loop's action on another loop.

[0036] This indicates the Hadamard product operation, which involves multiplying corresponding elements of two matrices of the same dimension one by one to obtain the relative gain array matrix. This operation is used to normalize the calculation of the relative gain ratio of each channel and eliminate the influence of dimensions.

[0037] Configure a temperature-coupled disturbance observer to estimate the unmodeled coupling disturbance between the two loops in real time, and convert the unmodeled coupling disturbance into an equivalent temperature disturbance feedforward to the corresponding control loop to compensate the residual of the decoupling compensator.

[0038] Furthermore, the thermal balance anomaly control module is based on the principle of temperature conservation. It calculates the total incident heat flow by integrating the surface temperature field, calculates the molten salt absorption power by combining the molten salt inlet and outlet temperatures with the operating flow rate, and simultaneously calculates the heat dissipation loss power to obtain the thermal balance imbalance index. It also sets three threshold levels for graded control. When the imbalance ratio is less than or equal to 10%, each loop executes the control strategy independently; when the imbalance ratio is between 10% and 20%, the monitoring sampling frequency is increased and abnormal data is recorded; when the imbalance ratio exceeds 20%, the CBF safety threshold for surface temperature control is tightened, the predictive feedforward compensation for outlet temperature is suspended, the adjustment authority for molten salt flow rate and heat flow rate is directly taken over, and the adjustment rate and range are limited.

[0039] Furthermore, the multi-objective arbitration module constructs a quadratic temperature objective function, which includes three sub-items: surface temperature uniformity index, outlet temperature stability index, and temperature change rate index. The priority of the three is adjusted by weighting coefficients. Establish a weight adjustment rule driven by temperature safety margin, calculate in real time the magnitude margin of the highest temperature from the material limit, the deviation margin of the outlet temperature from the set value, and the rate margin of the temperature change rate from the limit value, and tilt the weight towards the temperature index with the lowest margin. Under high irradiation conditions, the weight of outlet temperature control is increased; under medium irradiation conditions, equal weight is adopted; under low irradiation conditions, the weight of surface temperature control is increased; under extreme transient conditions, the weight of temperature amplitude and temperature change rate control is significantly increased, while the temperature gradient and adjustment rate are limited; a weighted smooth transition algorithm is adopted, and an upper limit is set for the weight change rate, so that the weight coefficients are continuously and smoothly adjusted when switching operating conditions.

[0040] This invention also provides a hierarchical coordinated control method for receivers in tower-type solar thermal power plants, based on the above-mentioned system, including the following specific steps: Step S0: Construct a quadratic temperature objective function that includes three types of indicators: surface temperature uniformity, outlet temperature stability, and temperature change rate; calculate the three types of temperature safety margins in real time and tilt the weights towards the low margin indicators to adapt to the priorities of different irradiation conditions; and limit the weight change rate through a weight smoothing transition algorithm. Step S1: The surface temperature data of the absorber is acquired using an infrared thermal imager array. The raw data is processed by non-uniformity correction, noise filtering and coordinate mapping. Initially, a basic partition is generated based on optical simulation. During operation, the partitions are dynamically refined or merged according to the local temperature gradient. The temperature characteristic parameters of each partition are output according to the effective partition. Step S2: Establish a linear state-space model of the absorber surface temperature, with the average temperature of the zone as the state vector and the heat flow regulation amount of the zone as the control input. The model parameters are corrected online by subspace identification and recursive least squares method. Construct a two-dimensional control barrier function constraint of temperature amplitude and temperature change rate, use intrinsic orthogonal decomposition to reduce the order, and output the heat flow regulation command after low-dimensional rolling optimization and verification. Step S3: Using a sequence-to-sequence architecture, a long short-term memory network is used to input historical normal direct irradiance and outlet temperature data, and output a multi-step irradiance prediction sequence. The prediction results are processed through a three-level mechanism of confidence interval estimation, error autoregression correction, and extreme weather degradation, and the heat flux conversion factor is corrected by combining the temperature deviation reverse correction channel. Step S4: A three-level architecture of prediction correction, phase feedforward, and fuzzy feedback is adopted. The residence time in the tube is calculated based on the molten salt temperature hysteresis model. The irradiation prediction sequence is aligned with the temperature hysteresis phase to generate the flow feedforward compensation amount. A piecewise adaptive fuzzy proportional-integral-derivative controller is used to adjust the parameters online. The feedforward and feedback amounts are superimposed to output the molten salt flow control command. Step S5: For each control cycle, the temperature relative gain array is calculated based on the dynamic gain matrix to quantify the coupling degree of the dual loop; the control mode is switched according to the coupling strength, and the decoupling compensation coefficient is continuously adjusted under medium and high coupling conditions. The loop cross-influence is offset by the decoupling compensator, and the residual is compensated in conjunction with the temperature coupling disturbance observer to synchronously constrain the single-cycle adjustment amplitude of the dual loop. Step S6: Calculate the incident heat flux, molten salt absorption power and heat dissipation loss power based on the principle of temperature conservation, generate the imbalance index and set three-level control thresholds; under severe imbalance conditions, tighten safety constraints, suspend the prediction feedforward compensation of outlet temperature, take over control authority and limit the adjustment rate, and gradually delegate control authority through a smooth transition mechanism after the anomaly is recovered.

[0041] The beneficial effects of this invention are as follows: 1. This invention employs a temperature gradient-driven adaptive partitioning acquisition strategy. Initially, it generates basic partitions with fine centers and coarse edges based on optical simulation. During operation, it dynamically refines or merges partitions according to local temperature gradients, achieving refined temperature measurement and control at the partition level. By constructing a linear state-space model with online parameter correction and combining intrinsic orthogonal decomposition to reduce the solution scale, it maintains constraint verification accuracy while ensuring solution efficiency. It is equipped with dual-dimensional control barrier function constraints for temperature amplitude and rate of change, sets two levels of safety thresholds, and dynamically adjusts convergence characteristics according to safety margin. It can effectively smooth out local temperature differences under high irradiance and cloud transient conditions, control surface temperature deviation within a narrow range, avoid local overheating and thermal fatigue damage, extend the service life of the receiver tube screen, and reduce the risk of overheating interlock shutdown.

[0042] 2. This invention constructs a sequence-to-sequence architecture Long Short-Term Memory network to achieve multi-step normal direct irradiance prediction, and is equipped with a three-level robust enhancement mechanism of confidence interval estimation, error autoregression correction, and extreme weather degradation, combined with temperature deviation reverse correction of heat flow conversion coefficient; the outlet temperature regulation adopts a three-level architecture of prediction correction, phase feedforward, and fuzzy feedback, and calculates the residence time in the pipe based on the molten salt flow temperature hysteresis model, accurately aligning the irradiance prediction sequence with the temperature hysteresis phase, and performing pre-regulation of flow rate in advance; the piecewise adaptive fuzzy proportional-integral-derivative controller matches an independent rule base according to the temperature deviation segment, and strengthens the differential suppression effect under over-limit conditions; it reduces the outlet temperature overshoot under DNI step fluctuations, maintains stable outlet temperature, and ensures continuous and stable operation of downstream thermal storage and power generation systems.

[0043] 3. This invention establishes a real-time monitoring mechanism for a dynamic relative gain array, quantifies the coupling strength of the dual loops of surface temperature uniformity and outlet temperature stability, and switches between three modes—distributed control, continuous decoupling, and centralized control—according to the degree of coupling. It also uses a temperature coupling disturbance observer to compensate for unmodeled residuals, eliminating dual-loop command conflicts and regulation oscillations. Based on the principle of temperature conservation, a three-level imbalance control mechanism is constructed, automatically tightening safety constraints, suspending predictive feedforward, and taking over control authority under abnormal operating conditions. A multi-objective arbitration module constructs a unified temperature objective function, dynamically adjusting the weights of three types of indicators based on temperature safety margin to adapt to the control priorities of different irradiation conditions, achieving collaborative optimization of multiple performance indicators while ensuring equipment safety. Attached Figure Description

[0044] Figure 1 This is a flowchart illustrating the overall closed-loop process of hierarchical coordinated control in this invention. Figure 2 This is a flowchart illustrating the internal execution of the temperature field prediction and control module of the present invention. Detailed Implementation

[0045] 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.

[0046] like Figures 1 to 2 As shown, this embodiment of the invention provides a hierarchical coordinated control system for the receiver of a tower solar thermal power plant, including a temperature field zone acquisition module, a temperature field prediction and control module, an irradiation prediction and correction module, an outlet temperature regulation module, a dual-loop decoupling module, a thermal balance anomaly control module, and a multi-objective arbitration module. This system is suitable for molten salt tower solar thermal power plants with capacities ranging from 50MW to 200MW. For 50MW power plants, it is recommended to have 40 to 80 control zones and a control cycle of 1500ms. For 100MW power plants, it is recommended to have 80 to 120 control zones and a control cycle of 500 to 1000ms. For 200MW power plants, it is recommended to have 120 to 200 control zones. A regional parallel computing strategy can be adopted to keep the total solution time within 1000ms.

[0047] In this embodiment of the invention, the temperature field zoning acquisition module uses an infrared thermal imager to form a planar array acquisition array to measure the temperature of the entire surface of the absorber. The resolution of a single device is not less than 640×512 pixels, the sampling frequency is 30Hz, and the spatial resolution is better than 0.15m×0.15m. An adaptive zoning strategy driven by temperature gradient is adopted. In the initial state, the heat flux density distribution on the surface of the receiver under different irradiation conditions is pre-calculated through optical simulation to generate a basic zoning scheme. The central high-energy flow area is divided into fine control zones, and the peripheral low-energy flow area is divided into coarse-grained control zones. The total number of zones is configured according to the power plant capacity. During operation, the temperature gradient values ​​of each region are calculated in real time. When the local temperature gradient exceeds the set threshold, the corresponding region is automatically subdivided. When the local temperature gradient remains below the set threshold, adjacent low-gradient regions are automatically merged. The total number of regions is adjusted within a preset range. The partitioning is refined by a two-part splitting method, dividing each target partition into two segments along the two orthogonal directions of the receiver's circumference and axis, forming four subdivided control units with equal area. The splitting boundary is aligned with the physical boundary of the receiver's tube array. For partition merging to be effective, the sliding average temperature gradient of adjacent partitions must be lower than a set threshold for five consecutive control cycles, and the average temperature difference between the two partitions must be less than a set limit. Merging is only performed on two partitions that share a complete adjacent boundary. Independent partitions at the edges are not included in cross-region merging. The adjustment range for the total number of partitions is set to 0.6 to 1.5 times the number of basic partitions. All refinement operations will stop when the upper limit of the range is reached, and all merging operations will stop when the lower limit of the range is reached.

[0048] The collected raw temperature data is processed sequentially through non-uniformity correction, noise filtering, and coordinate mapping. The average temperature, maximum temperature, and temperature gradient of each zone are statistically obtained according to the current effective zone and output to the temperature field prediction and control module.

[0049] Non-uniformity correction adopts the blackbody two-point calibration method, selecting two calibration temperature points below the lower limit of the working temperature and above the upper limit of the working temperature, and fitting the response correction coefficient pixel by pixel. Full-range calibration is performed when the equipment is turned on, and online single-point verification and correction are performed every 2 hours during operation. Noise filtering first uses 3×3 spatial median filtering to remove isolated abnormal pixels, and then uses 5th order temporal moving average filtering to smooth the temperature time-series fluctuations of single pixels. The coordinate mapping process, based on pre-calibrated thermal imager intrinsic and extrinsic parameters and distortion coefficients, projects the temperature data of the pixel plane onto the three-dimensional physical coordinate system of the absorber tube screen, and then maps it into two-dimensional unfolded plane coordinates to match the boundary coordinate range of each current partition.

[0050] In this embodiment of the invention, the temperature field prediction and control module establishes a linear state-space model of the absorber surface temperature, takes the average temperature of each zone as the state vector, and takes the zone heat flow adjustment amount corresponding to each group of heliostats as the control input. The initial state transition matrix and input matrix are obtained by linearizing the energy conservation equation at the rated operating point. During operation, the parameters of the linear state-space model are corrected online using the subspace identification method and the recursive least squares method to compensate for parameter drift caused by factors such as heliostat tracking error and mirror contamination. Online calibration takes the deviation between the measured temperature of each control cycle and the temperature predicted by the model in one step as input, and uses the recursive least squares method with a forgetting factor to incrementally update the coefficients of the state transition matrix and the input matrix. The forgetting factor is fixed at 0.98. When the relative deviation of the model prediction exceeds the set threshold for three consecutive control cycles, or when a step change in irradiance is detected that exceeds the set amplitude, the subspace identification method is activated to perform a full parameter batch update. After the update is completed, the covariance matrix of the recursive least squares method is reset. Under normal operating conditions, only incremental correction is performed. The above linear state-space model is valid within the following operating conditions: normal direct irradiance range of 400 to 900 W / m. 2 If the surface temperature of the absorber deviates from the rated value by less than ±50℃ and the mass flow rate of the molten salt deviates from the rated value by less than ±20%, the system will automatically switch to conservative control mode, reduce the control increment amplitude to 30% of the normal value, and re-trigger the full parameter identification program.

[0051] A control barrier function (CBF) safety constraint is constructed, defining two types of safety boundary functions for each partition: temperature amplitude and temperature change rate. The temperature amplitude is set with two levels of safety thresholds: the first threshold is 95% of the limit temperature of the tube screen material, and the second threshold is 92%. The temperature change rate is set with upper limit thresholds for the single-cycle temperature rise and fall rates. Both types of constraints adopt the form of exponential discrete CBF constraints, and the convergence time constant is adjusted according to the current temperature safety margin. The preferred value for the CBF attenuation coefficient is 0.05, which corresponds to a convergence time constant of approximately 20 seconds for the safety boundary function, balancing safety and control performance. When the attenuation coefficient is 0.01, the corresponding convergence time constant is approximately 100 seconds, and when it is 0.10, the corresponding convergence time constant is approximately 10 seconds. The value can be adjusted within the range of 0.01 to 0.10 according to the actual operating requirements of the power plant.

[0052] The temperature safety margin is calculated as the ratio of the difference between the highest temperature in the current zone and the limiting temperature of the tube-screen material to the limiting temperature. When the safety margin is lower than the first-level threshold, the convergence time constant is minimized, and the constraints are forcibly tightened. When the safety margin is between the first-level and second-level thresholds, the convergence time constant increases linearly with the safety margin. When the safety margin is higher than the second-level threshold, the convergence time constant is maximized, and the constraints remain relaxed. Constraint verification is performed in the order of temperature amplitude first, then temperature change rate. If the amplitude constraint is violated, correction is triggered directly, and the rate constraint is not verified.

[0053] Temperature field data under normal operating conditions is collected in a rolling manner to construct a snapshot matrix. The snapshot matrix is ​​updated at fixed intervals. The first m modes with an energy ratio of more than 95% are selected by singular value decomposition to construct a projection matrix. The intrinsic orthogonal decomposition (POD) technique is used to complete the mode order reduction, reduce the dimension of the original state space, and reduce the solution scale. Each column of the snapshot matrix corresponds to a full-zone temperature state vector at a sampling time. The sampling interval is consistent with the control cycle. Before construction, temperature data under abnormal operating conditions are removed, and only steady-state and normal transient operating condition samples are retained. The snapshot matrix is ​​fully updated every 24 hours, incorporating the latest normal operating condition temperature data and removing the oldest equivalent amount of historical data.

[0054] After singular value decomposition, modes are sorted from high to low according to their energy percentage. When the cumulative energy percentage reaches 95%, the corresponding order is truncated to construct the projection matrix. During the order reduction solution, the high-dimensional state vector is multiplied by the transpose of the projection matrix to obtain the low-dimensional state vector. After the optimization solution is completed, the low-dimensional control increment is multiplied by the projection matrix to map back to the original high-dimensional space, resulting in the control increment sequence of the entire partition.

[0055] Configurable regional POD reduction can be used as an alternative strategy. The receiver surface is divided into high-energy flow region, transition region and low-energy flow region according to the heat flux density distribution. Independent POD reduction models are established for each region to avoid reconstruction errors of a single global POD model in high temperature gradient regions.

[0056] The solution process adopts a collaborative mechanism of order reduction optimization and full-dimensional dual-dimensional verification. After obtaining the control increment by completing the rolling optimization solution of model predictive control in the low-dimensional space, the control increment is mapped back to the original high-dimensional space through the projection matrix. The two types of CBF constraints, namely temperature amplitude and temperature change rate, are verified point by point for all partitions. Model predictive control sets two types of parameters: prediction time domain and control time domain. The prediction time domain covers the next 10 control cycles, and the control time domain covers the next 3 control cycles. Optimization is performed only on the control increment within the control time domain, while the control increment for the remaining steps in the prediction time domain remains unchanged from its final value in the control time domain. A constraint on the rate of change of control increment is simultaneously incorporated into the optimization process to limit the difference in adjustment action between adjacent control cycles.

[0057] If a constraint violation occurs, the constraint projection method is used to correct the control increment to the feasible region boundary. Combined with a warm-start strategy and multi-threaded parallel computing, the single-cycle solution time of the partitioned system is shortened.

[0058] The output zone heat flux adjustment amount is mapped to the aiming point offset command of each group of heliostats through the heliostat aiming allocation algorithm. Each control zone corresponds to a fixed set of heliostats in advance. When a zone needs to increase the incident heat flux, the aiming points of the heliostats in the corresponding subset are gradually converged towards the geometric center of the zone, and the offset step size is proportionally converted according to the heat flux adjustment amount.

[0059] When a zone needs to reduce incident heat flux, the aiming point of some heliostats is radially shifted to the adjacent low-energy flux zone, or the corresponding heliostats are temporarily switched out of operation. During the adjustment process, the heliostats that are ranked later are adjusted in order of the amount of heat flux adjustment.

[0060] Before making corrections, all partitions are traversed to count the number of partitions that violate safety constraints and the amount of constraint violation for each partition. The partitions are then sorted from largest to smallest by the amount of violation. The control increment corresponding to the first partition in the sorting is adjusted first, and the heat flow regulation amplitude of the partition is compressed proportionally according to the constraint violation ratio. During the compression process, the constraint status of adjacent partitions is checked simultaneously. After a single partition is corrected, all partitions are retraced. If there are still constraint violations, the above sorting and correction process is repeated until all partitions meet the safety constraints. A fixed upper limit is set for the number of times a single correction loop is performed. When the upper limit is exceeded, a global uniform compression strategy is adopted.

[0061] In this embodiment of the invention, the irradiation prediction and correction module adopts a sequence-to-sequence architecture long short-term memory network to realize multi-step prediction of direct normal irradiance (DNI). The network includes two sub-networks: an encoder and a decoder. The encoder consists of two unidirectional long short-term memory (LSTM) network layers stacked together with a dropout layer. The number of hidden units in the first LSM network layer is set to 128, the number of hidden units in the second LSM network layer is set to 64, and the dropout layer is located between the two LSM network layers with a dropout rate of 0.2.

[0062] The encoder receives a historical input sequence of length 6 steps and finally outputs a context hidden state vector with a dimension of 64. The decoder consists of a long short-term memory network layer and a fully connected output layer. The number of hidden units in the long short-term memory network layer is the same as the encoder output dimension of 64. The number of neurons in the fully connected output layer is the same as the single-step prediction feature dimension. The decoder uses the context vector output by the encoder as the initial hidden state and gradually generates the normal direct irradiance prediction value for the next 10 steps. The encoder and decoder use an end-to-end sequence-to-sequence connection, and a teacher-forced strategy is used during the training phase to accelerate convergence.

[0063] The encoder inputs the DNI values ​​and exit temperature values ​​from 4 to 8 historical steps and extracts temporal features; the decoder outputs a DNI prediction sequence for the next 5 to 15 steps based on the hidden state output by the encoder, with a prediction duration covering 40 to 120 seconds. A three-level robustness enhancement mechanism is constructed. The first level is the estimation of the prediction confidence interval. During the inference phase, the Dropout layer is kept active. Predictive samples are generated through Monte Carlo sampling, and the prediction mean and 95% confidence interval are calculated. When the interval width exceeds a threshold, the lower bound value is used as the feedforward input. The confidence interval width threshold is set to 50 W / m. 2 .

[0064] The Monte Carlo sampling was set to 50 times. Each sampling randomly activated Dropout neurons at different locations in the network, generating a set of independent predicted sequence samples. All sampling results were sorted according to the prediction step size. The 2.5% quantile was used as the lower bound of the confidence interval, and the 97.5% quantile was used as the upper bound of the confidence interval. The ratio of the difference between the upper and lower bounds to the predicted mean was calculated as the interval width index.

[0065] When the width index exceeds the set threshold, the lower bound of the confidence interval is selected as the feedforward input for the corresponding step size; when it does not exceed the threshold, the predicted mean is used as the feedforward input.

[0066] The second stage is error autoregressive correction, which establishes a moving average model of historical prediction errors to estimate system deviations and correct prediction outputs in real time. The moving average model employs a fixed-length sliding window mechanism, with the window length set to 10 steps. Each step's sample corresponds to the same time lag step length in the prediction sequence. At each moment, the deviation between the measured normal direct irradiance of the current step and the corresponding historical prediction value is collected, and the error samples within the sliding window are updated. The arithmetic mean of the errors within the window is calculated as the system deviation estimate, and the deviation values ​​are superimposed point by point onto the prediction sequences of future steps. Before superposition, the deviation estimate is subjected to a first-order inertial filter, and the filter time constant is matched with the control period.

[0067] The third level is the extreme weather de-escalation strategy, which is implemented when a step change in DNI exceeding 200 W / m within a short period of time is detected. 2 When the predicted output is paused, the feedforward mode of the measured value is switched. The step change detection uses the difference between the measured normal direct irradiance of two consecutive sampling periods to determine the result. The sampling period is consistent with the irradiance acquisition period. After the degradation mode is triggered, the prediction feedforward channel is immediately cut off, and the current measured irradiance value is used as the feedforward input. At the same time, the prediction sequence buffer is cleared. The downgrade mode is maintained for a minimum duration of 30 seconds, during which the irradiance fluctuation amplitude is continuously monitored. When the single-cycle fluctuation amplitude falls below the threshold for 15 consecutive seconds, the predicted output ratio is gradually restored in a linearly increasing manner until it is completely switched back to the prediction feedforward mode.

[0068] A temperature deviation reverse correction channel is constructed to convert the deviation between the measured value of the outlet temperature and the calculated value of the heat balance model into an equivalent heat flux deviation, and the heat flux conversion factor corresponding to the DNI prediction sequence is corrected in real time. The corrected heat flux conversion factor is multiplied point by point with the normal direct irradiance prediction sequence to obtain the estimated value of the incident heat flux density of the receiver surface zone corresponding to each prediction step. The estimated value is directly used as the heat flux input of the phase feedforward link of the outlet temperature adjustment module, so as to realize the precise linkage between the irradiation prediction time sequence and the molten salt flow feedforward action.

[0069] The network training uses 12 consecutive months of operational data from the target power plant, with a total sample size of over 7 million groups. The Huber loss function is used, along with dynamic scheduling of the learning rate and an early stop strategy.

[0070] Before training, the raw operating data underwent standardization preprocessing: First, the Z-score standardization method was used to normalize the normal direct irradiance and outlet temperature data, respectively. The mean and standard deviation of the normalization were calculated based on the statistical values ​​of the training set. Then, a sliding time window was used to construct supervised learning samples. The window sliding step size was 1. The length of the input window corresponds to the number of historical input steps of the encoder, and the length of the output window corresponds to the number of prediction steps of the decoder.

[0071] The preprocessed dataset was randomly divided into training, validation, and test sets in an 8:1:1 ratio. The Adam optimizer was used during training, with an initial learning rate of 0.001. A validation loss plateau decay strategy was employed for learning rate scheduling; the learning rate was multiplied by 0.5 if the validation loss did not decrease for five consecutive rounds. An early stopping strategy was implemented, terminating training if the validation loss did not decrease for ten consecutive rounds, and saving the model weights with the lowest validation loss.

[0072] In this embodiment of the invention, the outlet temperature regulation module adopts a three-level architecture of prediction correction, phase feedforward, and fuzzy feedback. The first level is the prediction correction stage, in which the irradiation prediction correction module provides the future DNI change trend and heat flow conversion correction coefficient. The second level is the phase feedforward stage, which calculates the average residence time of molten salt in the absorber tube in real time based on the energy conservation equation and the molten salt flow temperature hysteresis model. It aligns the DNI prediction sequence with the molten salt temperature hysteresis phase, determines the lead amount and regulation rate of molten salt flow pre-regulation, and completes the flow pre-regulation synchronously with the heat flow change reaching the outlet section. The molten salt flow feedforward compensation amount is calculated based on the DNI prediction change. The molten salt flow temperature hysteresis model divides the receiver pipeline into three sections: the inlet rising section, the top heat exchange section, and the outlet falling section. The nominal residence time of each section is calculated based on the pipeline length and the real-time flow velocity of the molten salt, and the total residence time of the fluid is obtained by summing them up.

[0073] By combining the tube wall metal heat capacity parameters with the current molten salt flow range, the tube wall heat storage correction coefficient is selected by referring to the table, and the total residence time is corrected. Finally, the total lag time of heat flow change from the tube wall to the molten salt outlet is obtained. Based on this, the phase of the irradiation prediction sequence and the flow regulation action are aligned.

[0074] The third stage is a fuzzy feedback loop, which uses a piecewise adaptive fuzzy proportional-integral-derivative controller. The outlet temperature deviation and the rate of change of deviation are inputs, and the three PID parameters are adjusted online through fuzzy inference. Temperature deviation is divided into three segments: normal, warning, and over-limit. Each segment is configured with an independent fuzzy rule base and parameter adjustment range. In the over-limit segment, the differential action weight is forcibly increased. The absolute value of the outlet temperature deviation in the normal segment is less than or equal to 5℃, the absolute value of the deviation in the warning segment is between 5℃ and 15℃, and the absolute value of the deviation in the over-limit segment is greater than 15℃.

[0075] When switching sections, a hysteresis threshold is set. An increase in deviation triggers a section upgrade using the upper limit threshold, while a decrease in deviation triggers a section downgrade using the lower limit threshold. The difference between the upper and lower limits is set to 1°C to avoid frequent switching of the rule base at the boundary. The fuzzy inference system adopts a dual-input, three-output structure. The input variables, temperature deviation and deviation change rate, are divided into five fuzzy subsets. The output variables, proportional coefficient, integral coefficient, and differential coefficient, are continuously adjustable within a set range. The centroid method is used to complete the defuzzification calculation.

[0076] The core fuzzy rules follow the control logic of increasing proportional and derivative actions when the deviation is large and increasing integral actions when the deviation is small. The weight of derivative actions is forcibly increased in the over-limit section, and independent rule bases and parameter adjustment ranges are configured for different temperature deviation sections.

[0077] The universe of discourse for the input variable temperature deviation is set to -30℃ to 30℃, and the universe of discourse for the input variable deviation change rate is set to -5℃ / s to 5℃ / s. The five fuzzy subsets of the two input variables each correspond to five levels: negative large, negative small, zero, positive small, and positive large. Triangular membership functions are used for all subsets, while semi-triangular membership functions are used for the negative large and positive large subsets. The vertices of each subset are evenly distributed along the universe of discourse.

[0078] In the normal range, the output universe of discourse for the proportional coefficient is 0.8 to 2.5, for the integral coefficient it is 0.02 to 0.1, and for the differential coefficient it is 5 to 20. In the warning range, the output universe of discourse for the proportional coefficient is adjusted to 1.2 to 3.0, and for the differential coefficient it is adjusted to 10 to 30. In the over-limit range, the output universe of discourse for the proportional coefficient is adjusted to 1.5 to 3.5, for the differential coefficient it is adjusted to 15 to 40, and for the integral coefficient it remains unchanged.

[0079] The core fuzzy rule follows the control logic of increasing the proportional and derivative actions when the deviation is large and increasing the integral action when the deviation is small. For example, when the temperature deviation is large and the rate of change of the deviation is large, the proportional coefficient is large, the integral coefficient is small, and the derivative coefficient is large. When the temperature deviation is zero and the rate of change of the deviation is zero, the proportional coefficient is zero, the integral coefficient is positive, and the derivative coefficient is zero. The proportional, integral, and derivative parameters obtained by defuzzification are directly input to the proportional-integral-derivative controller of the molten salt flow control loop. After updating the controller parameters in real time, the final molten salt flow regulation command is generated.

[0080] In this embodiment of the invention, the dual-loop decoupling module establishes a real-time temperature dynamic relative gain array (RGA) monitoring mechanism. In each control cycle, based on the gain matrix of the control quantity to temperature under the current operating condition, the temperature RGA matrix is ​​calculated in real time using the relative gain array formula to quantify the coupling degree between the surface temperature uniformity control loop and the outlet temperature stability control loop. The gain matrix is ​​updated online using an incremental recursive identification method. In each control cycle, the incremental heat flow regulation and molten salt flow regulation of the previous cycle are collected, as well as the changes in the average surface temperature and the outlet temperature of the corresponding cycle. Using the control increment as input and the temperature change as output, the static gain coefficients of the four control channels are updated using a recursive least squares algorithm with a forgetting factor. The forgetting factor value is consistent with that of the model correction stage. After identification, the four gain coefficients are combined into a two-dimensional gain matrix according to the channel pairing relationship, which is used as the calculation input for the relative gain array.

[0081] When the main diagonal element of RGA is greater than or equal to 0.8, the pairing relationship between the main diagonal controlled variable and the control variable of surface temperature-heat flow regulation and outlet temperature-molten salt flow is maintained, a distributed control mode is adopted, and the decoupling compensation coefficient is set to zero. When the main diagonal element is in the range of 0.35 to 0.8, a continuous decoupling enhancement strategy is initiated. The decoupling compensation coefficient is adjusted linearly and continuously with the RGA value. The decoupling compensator matrix is ​​calculated based on the current RGA matrix. The temperature cross-coupling effect between the loops is offset by feedforward compensation. At the same time, a joint limiting constraint is applied to the control increment of the two loops to limit the adjustment amplitude of any channel within a single cycle. Within a single cycle, the adjustment amplitude of any channel is limited to no more than 15% of the rated value. The joint limiting adopts a combination of independent limiting of a single channel and joint limiting of the total adjustment amount. When the total adjustment amount exceeds the upper limit, it is compressed proportionally according to the control priority. The joint limiting method combines single-channel independent limiting with total regulation limiting. It sets single-channel increment upper limits for both the heat flow regulation loop and the molten salt flow loop, and sets a weighted sum upper limit for the regulation increments of the two loops.

[0082] When the total adjustment exceeds the joint upper limit, the control increment of each loop is compressed proportionally according to the current control priority of the two loops. The priority value is calculated by the corresponding weight coefficient of the multi-objective arbitration module.

[0083] When the main diagonal element is less than 0.35, the system switches to centralized control mode, where the multi-objective arbitration module generates heat flow regulation and molten salt flow commands in a unified manner to maximize the offsetting of loop coupling effects. Configure a temperature-coupled disturbance observer to estimate the unmodeled coupling disturbance between the two loops in real time, and convert the unmodeled coupling disturbance into an equivalent temperature disturbance feedforward to the corresponding control loop to compensate the residual of the decoupling compensator.

[0084] The temperature-coupled disturbance observer selects the nominal model obtained by linearization under rated operating conditions as the benchmark model, takes the heat flow regulation of the surface temperature control loop and the molten salt flow regulation of the outlet temperature control loop as inputs, and outputs the nominal temperature prediction values ​​of the two loops.

[0085] The difference between the measured temperature values ​​of the two loops and the predicted values ​​of the nominal model is used as the correction input. The unmodeled coupling disturbance not covered by the relative gain array is estimated by the nominal inverse model. After the estimation result is processed by the first-order low-pass filter, it is converted into the equivalent temperature deviation of the corresponding loop and superimposed on the temperature feedback input of the corresponding control loop to achieve feedforward compensation.

[0086] In this embodiment of the invention, the heat balance anomaly control module is based on the principle of temperature conservation. It calculates the total incident heat flux by integrating the surface temperature field. During the calculation, the average temperature of each zone and the working fluid temperature on the molten salt side are extracted first. Combined with the thermal conductivity of the pipe wall metal, the convective heat transfer coefficient inside the pipe and the pipe wall thickness parameters, the incident heat flux density of the outer wall surface of each zone is inferred from the steady-state heat transfer relationship of the pipe wall. Then read the actual light-receiving area value of each zone, multiply the heat flux density of each zone by the corresponding light-receiving area to obtain the incident heat power of the zone, and sum the incident heat power of all zones to obtain the total incident heat flow of the entire absorber. The molten salt absorption power is calculated by combining the inlet and outlet temperatures of the molten salt with the operating flow rate. The heat loss power is calculated simultaneously. The unbalanced power is obtained by subtracting the molten salt absorption power and the heat loss power from the total incident heat power. The ratio of the unbalanced power to the rated incident heat power is used as the quantitative basis to obtain the thermal balance unbalance index. Three-level thresholds are set for graded control. The heat loss power encompasses both convective and radiative heat loss. Convective heat loss is calculated based on the difference between the average surface temperature of the receiver and the ambient temperature, combined with the convective heat transfer coefficient corresponding to the real-time ambient wind speed. Radiative heat loss is calculated based on the average surface temperature of the receiver, the ambient temperature, and the emissivity coefficient of the tube screen surface.

[0087] When the imbalance ratio is less than or equal to 10%, each loop executes its control strategy independently. When the imbalance ratio is between 10% and 20%, the monitoring sampling frequency is increased and abnormal data is recorded. When the imbalance ratio exceeds 20%, the CBF safety threshold for surface temperature control is tightened, the predictive feedforward compensation for outlet temperature is suspended, and the adjustment authority for molten salt flow rate and heat flow rate is directly taken over, with the adjustment rate and range limited. Specifically, the CBF primary safety threshold is temporarily tightened to 90% of the limit temperature of the tube screen material, the prediction time domain for temperature field prediction control is shortened to 2 / 3 of the normal value, the molten salt flow rate adjustment rate is limited to 50% of the normal value, and the heliostat aiming point adjustment range is limited to 60% of the normal range. During abnormal takeover, the multi-objective arbitration module will increase the temperature safety margin weight to the highest level, suspend performance index optimization, and prioritize equipment safety; the unbalance ratio is the proportion of temperature conservation unbalance to rated incident heat power. The intervention state is set with a smooth transition exit mechanism. When the temperature conservation imbalance returns to the normal range and remains stable for more than 3 minutes, the control is gradually delegated to the underlying control module.

[0088] The decentralization of control is divided into three consecutive phases. In the first phase, the rate restrictions on the regulation of molten salt flow and heat flow are gradually lifted, and the upper limit of the regulation rate is gradually increased to the normal value according to the control cycle.

[0089] The second phase involves restoring the export temperature prediction feedforward compensation, gradually increasing the feedforward compensation amount proportionally, and simultaneously gradually relaxing the safety threshold for surface temperature control to the normal value.

[0090] The third stage completely removes the abnormal control constraints, and each control loop resumes independent operation. Each stage lasts for no less than 1 minute. Before switching stages, it must be confirmed that the imbalance degree is continuously maintained within the normal range. If the imbalance degree rebounds to the abnormal range during the transition, it must immediately revert to the previous control stage.

[0091] In this embodiment of the invention, the multi-objective arbitration module constructs a quadratic temperature objective function, which includes three sub-items: surface temperature uniformity index, outlet temperature stability index, and temperature change rate index. The priority of the three is adjusted by weighting coefficients. The surface temperature uniformity index is represented by the square mean of the temperature deviations of each zone, the outlet temperature stability index is represented by the square of the outlet temperature deviation, and the temperature change rate index is represented by the maximum value of the temperature change in a single cycle of each zone; the sum of the weight coefficients of the three types of indices is 1, and the weight of a single index ranges from 0.3 to 0.7. The surface temperature uniformity index is characterized by the standard deviation of the average temperature of all zones, the outlet temperature stability index is characterized by the absolute value of the deviation between the measured outlet temperature and the set value, and the temperature change rate index is characterized by the maximum value of the temperature change in a single cycle for each zone.

[0092] All three types of indicators are normalized based on the corresponding benchmark values ​​under rated operating conditions and then incorporated into the objective function. The frequency of indicator calculation is synchronized with the system control cycle.

[0093] Establish a weight adjustment rule driven by temperature safety margin, calculate in real time the magnitude margin of the highest temperature from the material limit, the deviation margin of the outlet temperature from the set value, and the rate margin of the temperature change rate from the limit value, and tilt the weight towards the temperature index with the lowest margin. Before weight adjustment, the difference between the current actual value and the corresponding limit value of the three types of indicators is calculated respectively. The ratio of the difference to the limit benchmark value is used as the margin ratio. The margin ratio of the three types of indicators is normalized after taking the reciprocal, and the preliminary target weight allocation result is obtained. The preliminary weight result is used as the target weight input value and sent to the weight smooth transition algorithm for gradual update. The weight and verification are performed simultaneously during the update process to ensure that the sum of the weights of the three types of indicators remains constant.

[0094] DNI≥700W / m 2 For high-irradiation conditions, 300~700W / m 2 For medium irradiation conditions, <300W / m 2 For low irradiation conditions; for high irradiation conditions, increase the weight of outlet temperature control; for medium irradiation conditions, use equal weights; for low irradiation conditions, increase the weight of surface temperature control; for extreme transient conditions, significantly increase the weights of temperature amplitude and temperature change rate control, and limit temperature gradient and adjustment rate; adopt a weighted smooth transition algorithm, set an upper limit for the weight change rate, so that the weight coefficients are continuously and smoothly adjusted when switching operating conditions.

[0095] Within a single control cycle, the weight adjustment of any indicator shall not exceed 5% of the total weight. When the difference between the calculated target weight and the current weight exceeds this range, the target weight shall be gradually approximated by the upper limit. After each weight adjustment, a weight and verification process shall be performed to normalize the weights of the three types of indicators.

[0096] Extreme transient conditions are determined by combining the single-cycle change in normal direct irradiance and the rate of increase in the maximum surface temperature. When the single-cycle change in irradiance exceeds a set threshold, or the rate of increase in the maximum surface temperature exceeds a set limit, it is determined to be an extreme transient condition. The condition determination result remains valid until both indicators fall below the threshold for three consecutive control cycles, at which point the extreme transient condition weighting strategy is terminated.

[0097] This invention also provides a hierarchical coordinated control method for receivers in tower-type solar thermal power plants, based on the above-described system, including the following specific steps: Step S0: Construct a quadratic temperature objective function that includes three types of indicators: surface temperature uniformity, outlet temperature stability, and temperature change rate; calculate the three types of temperature safety margins in real time and tilt the weights towards the low margin indicators to adapt to the priorities of different irradiation conditions; and limit the weight change rate through a weight smoothing transition algorithm. Step S1: The surface temperature data of the absorber is acquired using an infrared thermal imager array. The raw data is processed by non-uniformity correction, noise filtering and coordinate mapping. Initially, a basic partition is generated based on optical simulation. During operation, the partitions are dynamically refined or merged according to the local temperature gradient. The temperature characteristic parameters of each partition are output according to the effective partition. Step S2: Establish a linear state-space model of the absorber surface temperature, with the average temperature of the zone as the state vector and the heat flow regulation amount of the zone as the control input. The model parameters are corrected online by subspace identification and recursive least squares method. Construct a two-dimensional control barrier function constraint of temperature amplitude and temperature change rate, use intrinsic orthogonal decomposition to reduce the order, and output the heat flow regulation command after low-dimensional rolling optimization and verification. Step S3: Using a sequence-to-sequence architecture, a long short-term memory network is used to input historical normal direct irradiance and outlet temperature data, and output a multi-step irradiance prediction sequence. The prediction results are processed through a three-level mechanism of confidence interval estimation, error autoregression correction, and extreme weather degradation, and the heat flux conversion factor is corrected by combining the temperature deviation reverse correction channel. Step S4: A three-level architecture of prediction correction, phase feedforward, and fuzzy feedback is adopted. The residence time in the tube is calculated based on the molten salt temperature hysteresis model. The irradiation prediction sequence is aligned with the temperature hysteresis phase to generate the flow feedforward compensation amount. A piecewise adaptive fuzzy proportional-integral-derivative controller is used to adjust the parameters online. The feedforward and feedback amounts are superimposed to output the molten salt flow control command. Step S5: For each control cycle, the temperature relative gain array is calculated based on the dynamic gain matrix to quantify the coupling degree of the dual loop; the control mode is switched according to the coupling strength, and the decoupling compensation coefficient is continuously adjusted under medium and high coupling conditions. The loop cross-influence is offset by the decoupling compensator, and the residual is compensated in conjunction with the temperature coupling disturbance observer to synchronously constrain the single-cycle adjustment amplitude of the dual loop. Step S6: Calculate the incident heat flux, molten salt absorption power and heat dissipation loss power based on the principle of temperature conservation, generate the imbalance index and set three-level control thresholds; under severe imbalance conditions, tighten safety constraints, suspend the prediction feedforward compensation of outlet temperature, take over control authority and limit the adjustment rate, and gradually delegate control authority through a smooth transition mechanism after the anomaly is recovered.

[0098] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0099] 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. A layered coordinated control system for the receiver of a tower-type solar thermal power plant, characterized in that, Includes the following modules: The temperature field zoning acquisition module performs infrared temperature measurement on the surface of the absorber, adaptively adjusts the granularity of the zoning based on the temperature gradient, and outputs the temperature characteristic parameters of each zone after correcting and filtering the original temperature measurement data. The temperature field prediction and control module establishes a state-space model of the absorber surface temperature and corrects the model parameters online. After reducing the optimization solution scale through modal order reduction, it rolls the solution to solve the control increment. Based on the control barrier function, it constructs a dual safety constraint of temperature amplitude and temperature change rate. Under the dual safety constraint, it verifies and corrects the output of the zoned heat flow regulation amount. The irradiance prediction and correction module is based on a temporal deep learning network to achieve multi-step prediction of normal direct irradiance. It improves the reliability of prediction through a robustness enhancement mechanism and combines temperature deviation to inversely correct the heat flux conversion factor. The outlet temperature regulation module adopts a multi-level control architecture. Based on the irradiation prediction results and the molten salt flow temperature hysteresis characteristics, it completes phase alignment and flow pre-regulation. After correction by the adaptive feedback controller, it outputs molten salt flow control commands. The dual-loop decoupling module quantifies the coupling degree between the two control loops of surface temperature uniformity and outlet temperature stability, adjusts the decoupling compensation strength according to the coupling strength, and compensates for the unmodeled coupling residuals in conjunction with the temperature coupling disturbance observer. The thermal balance anomaly control module calculates the imbalance index in real time and manages it in stages based on the principle of temperature conservation. Under abnormal operating conditions, it tightens safety constraints and takes over control authority, and configures a smooth transition exit mechanism. The multi-objective arbitration module constructs a quadratic temperature objective function, adjusts the weights of each index based on the temperature safety margin of each zone, adapts to different irradiation conditions and transient conditions, and optimizes control performance.

2. The layered coordinated control system for the receiver of a tower-type solar thermal power plant according to claim 1, characterized in that, The temperature field zoning acquisition module uses an infrared thermal imager to form a planar array acquisition array to measure the temperature of the absorber surface. The temperature gradient-driven adaptive zoning strategy is adopted. In the initial state, the heat flux density distribution of the absorber surface under different irradiation conditions is pre-calculated through optical simulation to generate a basic zoning scheme with fine high-energy flow zone in the center and coarse low-energy flow zone at the edge. During operation, the temperature gradient of each region is calculated in real time. When the local temperature gradient exceeds the threshold, the corresponding partition is refined. When the local temperature gradient is continuously below the threshold, adjacent low-gradient partitions are merged. The total number of partitions is adjusted within a preset range. The raw temperature measurement data is sequentially corrected, filtered, and mapped, and then the average temperature, maximum temperature, and temperature gradient parameters are statistically output according to each effective zone.

3. The layered coordinated control system for the receiver of a tower-type solar thermal power plant according to claim 2, characterized in that, The temperature field prediction and control module establishes a linear state-space model of the absorber surface temperature, with the average temperature of each zone as the state vector and the zone heat flow adjustment amount corresponding to each group of heliostats as the control input. The initial state transition matrix and input matrix are obtained by linearizing the energy conservation equation at the rated operating point. During operation, the subspace identification method and recursive least squares method are used to correct the model parameters online and compensate for parameter drift. Temperature field data is collected periodically to construct a snapshot matrix and the snapshot matrix is ​​updated. The principal mode is selected by singular value decomposition to construct a projection matrix. The mode order is reduced by using eigenvalue decomposition technology, thereby reducing the dimension of the original state space and the scale of the optimization solution.

4. The layered coordinated control system for the receiver of a tower-type solar thermal power plant according to claim 3, characterized in that, The temperature field prediction and control module constructs two types of control barrier function safety constraints for each partition: the temperature amplitude and the temperature change rate. The temperature amplitude is set with two levels of safety thresholds, and the temperature change rate is set with upper limits for single-cycle temperature rise and temperature fall rates. The convergence time constants of the two types of safety constraints are adjusted according to the current temperature safety margin of the partition. The solution process employs a collaborative mechanism of order reduction optimization and full-dimensional verification. After obtaining the control increment by completing the model predictive control rolling optimization in the low-dimensional space, the control is mapped back to the original high-dimensional space to verify the safety constraints point by point. If a constraint violation occurs, the control increment is corrected to the feasible region boundary, and the single-cycle solution time is shortened by combining a hot-start strategy and parallel computing technology.

5. The layered coordinated control system for the receiver of a tower-type solar thermal power plant according to claim 4, characterized in that, The irradiation prediction and correction module adopts a sequence-to-sequence architecture long short-term memory network to realize multi-step prediction of normal direct irradiance. The long short-term memory network includes two sub-networks: an encoder and a decoder. The encoder extracts temporal features by taking historical multi-step normal direct irradiance and exit temperature as input, and the decoder outputs a future multi-step prediction sequence based on the hidden state output by the encoder. The three-level robustness enhancement mechanism is constructed, consisting of prediction confidence interval estimation, error autoregressive correction, and extreme weather degradation strategy, to improve prediction reliability. A temperature deviation reverse correction channel is constructed to convert the deviation between the measured and calculated outlet temperature into an equivalent heat flow deviation, and the heat flow conversion factor is corrected in real time.

6. The layered coordinated control system for the receiver of a tower-type solar thermal power plant according to claim 5, characterized in that, The outlet temperature regulation module adopts a three-level control architecture of prediction correction, phase feedforward, and fuzzy feedback; the prediction correction stage receives the irradiation prediction result and the correction value of the heat flux conversion coefficient as input; The phase feedforward link is based on the energy conservation equation and the molten salt flow temperature hysteresis model. It calculates the average residence time in the molten salt tube, aligns the irradiation prediction sequence with the temperature hysteresis phase, and generates the molten salt flow feedforward compensation amount. The fuzzy feedback loop adopts a piecewise adaptive fuzzy proportional-integral-derivative controller, with the outlet temperature deviation and the deviation change rate as inputs for online adjustment parameters. Independent rule bases and parameter adjustment ranges are configured for different deviation segments. The molten salt flow control command is output after the feedforward compensation and feedback control quantities are superimposed.

7. The layered coordinated control system for the receiver of a tower-type solar thermal power plant according to claim 6, characterized in that, The dual-loop decoupling module establishes a real-time temperature relative gain array monitoring mechanism. In each control cycle, the relative gain array is calculated based on the gain matrix of the control quantity to temperature under the current operating condition, thereby quantifying the coupling degree between the two control loops. When the minimum value of the main diagonal element of the relative gain array is ≥0.8, a distributed control mode is adopted and the decoupling compensation coefficient is set to zero; when the minimum value of the main diagonal element of the relative gain array is in the range of 0.35~0.8, a continuous decoupling strategy is started, and the compensation coefficient is adjusted linearly with the coupling strength. The cross-coupling effect between loops is offset by feedforward compensation, and joint amplitude limiting constraints are applied to the dual-loop control increment simultaneously. When the minimum value of the main diagonal element of the relative gain array is less than 0.35, the control mode is switched to centralized control mode, and the control command is uniformly generated by the multi-objective arbitration module. Configure the temperature-coupled disturbance observer to estimate the unmodeled coupled disturbance and feed it forward to the corresponding control loop to compensate for the unmodeled coupled residual.

8. The layered coordinated control system for the receiver of a tower-type solar thermal power plant according to claim 7, characterized in that, The thermal balance anomaly control module calculates the imbalance index in real time and manages it in stages based on the temperature conservation principle. Under abnormal operating conditions, it tightens the safety constraints and takes over the control authority, and configures the smooth transition exit mechanism. Based on the temperature conservation principle, the total incident heat flow is calculated by integrating the surface temperature field. The molten salt absorption power and heat dissipation loss power are calculated by combining the molten salt inlet and outlet temperatures and the operating flow rate. The thermal balance imbalance index is obtained, and a three-level threshold graded control is set. When the imbalance is in the low range, each loop is controlled independently; in the middle range, the sampling frequency is increased and abnormal data is recorded; in the high range, the safety threshold for surface temperature control is tightened, the prediction feedforward compensation of the outlet temperature is suspended, control authority is taken over, and the adjustment rate and range are limited.

9. The layered coordinated control system for the receiver of a tower-type solar thermal power plant according to claim 8, characterized in that, The multi-objective arbitration module constructs a quadratic temperature objective function that includes three types of indicators: surface temperature uniformity, outlet temperature stability, and temperature change rate, and adjusts the priority of the three indicators through weighting coefficients. Establish a weight adjustment rule driven by the temperature safety margin of the partition, calculate the temperature amplitude margin, outlet temperature deviation margin and temperature change rate margin in real time, and tilt the weight towards the index with the lowest margin. To adapt corresponding weighting strategies to different irradiation conditions, the weight of safety indicators is increased under extreme transient conditions. A weighted smooth transition algorithm is adopted, and an upper limit is set for the weight change rate to achieve continuous and smooth adjustment of the weight coefficient when switching operating conditions.

10. A method for hierarchical coordinated control of receivers in a tower-type solar thermal power plant, based on the system described in any one of claims 1-9, characterized in that, The specific steps include the following: A quadratic temperature objective function containing three types of indicators is constructed. The weights of the indicators are dynamically adjusted based on the real-time temperature safety margin. The rate of change of the weights is limited by a weight smoothing transition algorithm to adapt to different irradiation conditions. The surface temperature data of the absorber is acquired by an area array infrared thermometry method. After correction, filtering and coordinate mapping preprocessing, the partitions are dynamically refined or merged according to the local temperature gradient, and the partition temperature characteristic parameters are output according to the effective partitions. A multi-step irradiation prediction sequence is generated through a temporal deep learning network. After being processed by a three-level robustness enhancement mechanism, the heat flux conversion factor is corrected in reverse by combining temperature deviation. A linear state-space model of the absorber surface temperature is established and the model parameters are corrected online. After modal order reduction, the model is solved by rolling optimization under the dual safety constraints of temperature amplitude and temperature change rate to generate the zoned heat flow regulation. Based on the molten salt flow temperature hysteresis model, the irradiation prediction phase is aligned to generate the molten salt flow feedforward compensation quantity, which is then combined with piecewise adaptive fuzzy feedback control to generate the molten salt flow control quantity. The coupling degree of the dual control loops is calculated in real time by using a relative gain array. The control mode is switched and the decoupling compensation strength is adjusted according to the coupling strength. The coupling residual is compensated by a temperature coupling disturbance observer, and the single-cycle adjustment amplitude is constrained to obtain the decoupled control increment. Based on the principle of temperature conservation, the thermal balance imbalance is calculated in real time and graded control is carried out according to three threshold levels. Under severe abnormal conditions, safety constraints are tightened and control authority is taken over. After the abnormality is recovered, control authority is gradually delegated through a smooth transition exit mechanism. The final zone heat flow regulation command and molten salt flow control command are output.

Images