CN-122015565-A - Dynamic compensation method for evaporation loss of cooling tower

CN122015565ACN 122015565 ACN122015565 ACN 122015565ACN-122015565-A

Abstract

The invention provides a dynamic compensation method for evaporation loss of a cooling tower, which is characterized in that key parameters such as water inlet and outlet temperature difference, circulating water quantity, environment wet bulb temperature, fan state, water supplementing record and the like are acquired in real time through an industrial sensor network and a historical database, are input into a high-level strategy network after standardized and normalized treatment, are dynamically generated into water-saving, energy-saving and stable priority weights and are modulated into a reward function structure, an optimal compensation water quantity instruction is deduced based on a depth deterministic strategy gradient algorithm by combining a multi-target weight self-adaptive adjustment mechanism, closed-loop dynamic compensation for the evaporation loss is realized, a cross-working condition migration training and on-line fine adjustment mechanism is introduced, and model generalization and on-site adaptation capability are improved.

Inventors

Shan Wenxian
Shan Wenqiao
LIU FENGJUN

Assignees

广州单梁全钢冷却塔设备有限公司

Dates

Publication Date: 20260512
Application Date: 20260202

Claims (10)

1. The dynamic compensation method for the evaporation loss of the cooling tower is characterized by comprising the following steps of: s1, acquiring real-time environmental parameters in the operation process of a cooling tower, wherein the real-time environmental parameters comprise inlet/outlet water temperature difference of the cooling tower, circulating water quantity, environmental wet bulb temperature, fan operation state and historical water supplementing record; S2, normalizing and standardizing the real-time environment parameters to generate standardized environment parameters; s3, inputting the standardized environmental parameters into a high-level strategy network, and generating a multi-target priority vector under the current working condition by combining a system operation target weight configuration file, wherein the multi-target priority vector comprises a water-saving priority mode, an energy-saving priority mode or a stable priority mode; S4, dynamically adjusting a reward function structure of the bottom action network based on the multi-target priority vector as a context modulation signal; s5, adopting a depth deterministic strategy gradient algorithm, taking a multivariable input state space as input, executing strategy network reasoning, and outputting a specific compensation water quantity adjusting instruction; S6, constructing a target competition degree evaluation unit, calculating Pareto front distance among targets, and triggering a weight rebalancing mechanism if a certain target is detected to continuously press other targets to exceed a set threshold value; S7, introducing a cross-working condition migration training strategy, and pre-training by using historical operation data from multiple power plants and multiple climate zones in an off-line training stage; s8, executing an online fine tuning mechanism in the running process, and updating local parameters of the pre-training model based on real-time feedback data of the current project; and S9, transmitting the generated compensation water quantity adjusting instruction to a dynamic compensation module.
2. The method for dynamic compensation of evaporative loss of cooling tower according to claim 1, wherein said step S9 further comprises: And S10, periodically evaluating long-term operation efficiency feedback, and dynamically evolving each target weight coefficient in the reward function to form a closed-loop self-adaptive optimization mechanism.
3. The method for dynamic compensation of evaporative loss of cooling tower according to claim 1, wherein said step S1 specifically comprises: Based on an industrial sensor network, collecting difference data of inlet water temperature and outlet water temperature in the operation process of the cooling tower, and obtaining a temperature gradient change rate; reading an operation state signal of the circulating water pump in real time, acquiring current circulating water quantity data, and constructing water quantity change trend characteristics; acquiring wet bulb temperature values of air around the cooling tower through an ambient temperature and humidity sensor, and calculating air saturation and evaporation potential; Based on a fan running state monitoring module, acquiring a fan start-stop state and a rotating speed feedback signal, and identifying the forced ventilation intensity of the current cooling tower; and (5) calling a historical operation record database of the water replenishing system, and extracting the water replenishing frequency and the total water replenishing amount in unit time to form historical water replenishing behavior characteristics.
4. The method of claim 3, wherein the step S1 further comprises synchronously collecting the inlet water temperature and the outlet water temperature based on an industrial sensor network, using a thermal resistance sensor Pt100, a precision class a, a sampling frequency of 1Hz, performing signal time alignment by using an IEEE1588 time synchronization protocol, performing data smoothing by using differential sampling, sliding window average filtering (window length 5min, overlapping rate 50%), and second order Butterworth low-pass filtering (cut-off frequency 0.005 Hz), and performing Z-score normalization processing to obtain a normalized temperature gradient feature vector.
5. The method for dynamic compensation of evaporative loss of cooling tower according to claim 1, wherein said step S2 specifically comprises: carrying out missing value detection and abnormal value elimination processing on the original environment parameters to obtain a complete and credible initial data set; performing standardization processing on the environmental parameters of each dimension based on a standardization method, and outputting standardized environmental parameter vectors; Performing unified interval mapping on discrete variables by adopting a minimum-maximum normalization method to form a normalization state vector; Performing feature scaling and dimension alignment operations on the normalized state vector to generate a normalized input tensor; And caching the standardized input tensor into a runtime data pool, and introducing Gaussian disturbance samples through a data enhancement mechanism.
6. The method for dynamic compensation of evaporative loss of cooling tower according to claim 1, wherein said step S3 specifically comprises: Performing feature coding operation on the standardized cooling tower environment parameters, performing nonlinear mapping on input feature vectors by adopting a fully connected neural network, extracting abstract feature representation suitable for high-level policy decision, and outputting feature embedding vectors; Based on the feature embedded vector, combining preset priority distribution parameters in a system operation target weight configuration file, executing target weight mapping operation by adopting a Softmax normalization function, generating an initial multi-target priority distribution vector, and outputting initial weight distribution coefficients of three targets of water saving, energy saving and stability; performing dynamic modulation enhancement operation on the initial multi-target priority level distribution vector, introducing a gating mechanism to perform nonlinear adjustment on each target weight, and outputting a modulated multi-target priority level vector; Performing a context signal coding operation based on the modulated multi-target priority vector, coding the multi-target priority vector into a context modulation signal by adopting a vector splicing technology, and outputting a context regulation vector; and transmitting the context regulation vector to a target competition degree evaluation unit, and outputting a priority evaluation factor.
7. The method of claim 6, wherein the step S3 further comprises obtaining an abstract feature embedded vector through feature encoding, combining preset weight distribution parameters, distributing three kinds of preliminary weights of targets of water saving, energy saving and stability through Softmax normalization mapping, and enabling the weight distribution to meet normalization constraint by utilizing parameter temperature adjustment and numerical precision correction algorithm.
8. The method for dynamic compensation of evaporative loss of cooling tower according to claim 1, wherein said step S4 specifically comprises: Weighting the water-saving target sub-rewarding function based on the water-saving priority component in the multi-target priority distribution vector; weighting the energy-saving target sub-rewarding function based on the energy-saving priority components in the multi-target priority vector; weighting the stability target sub-bonus function based on the stability priority component in the multi-target priority vector; Combining the weighted water-saving target sub-rewarding function, the energy-saving target sub-rewarding function and the stability target sub-rewarding function in a linear weighting mode to generate a comprehensive element rewarding function; and carrying out normalization processing on the meta-rewarding function to generate a normalized meta-rewarding function.
9. The method for dynamic compensation of evaporative loss of cooling tower according to claim 1, wherein said step S5 specifically comprises: Performing feature selection and dimension alignment processing on the multivariable data in the input state space to obtain a unified dimension standardized state vector; Inputting the standardized state vector into an action network of a depth deterministic strategy gradient algorithm, executing estimation calculation of an action cost function based on a rewarding function structure modulated by a current multi-target priority vector, and generating an optimal action strategy in a current state; Iterative optimization is carried out on the action network parameters through a strategy gradient updating mechanism, and a compensation water quantity adjusting instruction output by actions is adjusted by combining feedback signals of the element rewarding function; Based on an experience playback mechanism, sampling and training historical strategy execution data of an action network, and adopting a target network to perform stable estimation on an action cost function; And generating a final compensation water quantity adjusting instruction, wherein the instruction is a continuous control signal and represents a water quantity value required to be supplemented in unit time, and the water quantity value is used as a control input of the dynamic compensation module.
10. The method of claim 9, wherein the multivariate data comprises a rate of change of temperature gradient, a trend of liquid level fluctuation, and a strength of load fluctuation.

Description

Dynamic compensation method for evaporation loss of cooling tower Technical Field The invention relates to the technical field of multi-target collaborative optimization control of cooling towers, in particular to a dynamic compensation method for evaporation loss of a cooling tower. Background In the field of intelligent control of the operation of the cooling tower and multi-objective optimization of industrial processes, aiming at dynamic compensation of evaporation loss of the cooling tower, a main stream technical scheme generally comprises a rule-driven water quantity compensation method, a traditional prediction model (such as statistical regression and a simple neural network) and a reinforcement learning strategy based on single-objective optimization. The schemes aim at realizing high-efficiency water supplementing and energy consumption control of the cooling tower so as to improve the economical efficiency and the stability of the system operation; The prior art focuses on generating a water replenishing instruction by adopting a preset rule or a static rewarding mechanism according to the evaporation quantity prediction result, or weighting targets such as water saving, energy saving, system stability and the like by using experience parameters to finish the output of a compensation strategy. Some recent technologies begin to introduce reinforcement learning frameworks that promote accurate adjustment levels of the amount of compensation water through real-time sensing and strategic adaptive optimization of operating environment parameters. For example, the part of the reinforcement learning compensation system can automatically adjust the water supplementing amount according to the water inlet and outlet temperature difference, the circulating water amount and the environmental meteorological conditions of the cooling tower, and takes energy consumption management and system safety into consideration, so that the reinforcement learning compensation system becomes a development trend of the field; however, the current technology generally has the following limitations: (1) The compensation strategy adopts a single-target or static weighted rewarding mechanism, so that multi-target conflict of water saving, energy saving and stable system operation is difficult to consider, unbalance of resource allocation is easy to cause, and a certain target of water quantity, energy consumption or system stability is ignored for a long time; (2) When the running environment fluctuates severely, the traditional model cannot autonomously adjust the weight among the targets, and the compensation strategy is easy to fail or the running of the system is unstable. In addition, the existing reinforcement learning method lacks recognition and balance measures for competition and synergy among multiple targets, and risks of strategy collapse and bias optimization exist; (3) Most schemes rely on solidification rules or static parameter configuration, have limited generalization capability, and are difficult to adapt to dynamic changes under complex working conditions of different power plants, different climate areas and the like, so that the water resource management efficiency and the comprehensive energy efficiency of the system are improved. Disclosure of Invention The invention provides a dynamic compensation method for evaporation loss of a cooling tower in order to solve the technical problems. The technical scheme of the invention is realized by a dynamic compensation method for evaporation loss of a cooling tower, which comprises the following steps: S1, acquiring real-time environmental parameters in the running process of a cooling tower, wherein the real-time environmental parameters comprise inlet/outlet water temperature difference, circulating water quantity, environmental wet bulb temperature, fan running state and historical water supplementing record of the cooling tower, and the real-time environmental parameters are used as input conditions for multi-target collaborative optimization decision-making; S2, carrying out normalization and standardization processing on the acquired real-time environment parameters to eliminate interference of different dimension lines on a model training process and improve convergence efficiency of a subsequent reinforcement learning algorithm; S3, inputting the environment parameters subjected to standardized processing into a high-level strategy network, and generating a multi-target priority vector under the current working condition by combining a system operation target weight configuration file, wherein the multi-target priority vector comprises a water-saving priority mode, an energy-saving priority mode or a stable priority mode; S4, dynamically adjusting a reward function structure of the bottom action network based on the multi-target priority vector as a context modulation signal; S5, adopting a depth deterministic strategy gradient algorithm, taking a multi