A control method for coordinated dynamic balance of pressure differentials in multi-zone FFUs

Abstract

This invention discloses a control method for dynamic balance of pressure differential in multi-zone FFUs, belonging to the field of industrial automation and intelligent control. It includes: real-time acquisition of FFU operating parameters, environmental parameters, and energy consumption data in multiple cleanrooms; construction of an energy consumption assessment model and a multi-dimensional prediction model based on historical data; generation of an FFU group collaborative control strategy based on predicted energy consumption data, pressure differential change trends, and personnel flow change trends; issuance of dynamic adjustment commands; and continuous parameter acquisition for adaptive optimization of the model to ensure continuous dynamic balance of pressure differential in multi-zone FFUs. This invention employs predictive control and multi-dimensional data fusion technology, enabling intelligent group collaborative control and energy consumption optimization of FFUs, significantly improving the pressure differential stability and operating efficiency of cleanrooms.

Description

Technical Field

[0001] This invention relates to the field of industrial automation and intelligent control, and in particular to a control method for coordinated dynamic balancing of pressure differentials in multi-region FFUs. Background Technology

[0002] Fan filter units (FFUs) are critical devices in cleanrooms used to maintain air cleanliness and specific pressure differential gradients. They are widely used in industries with extremely high environmental requirements, such as semiconductor manufacturing, biopharmaceuticals, and precision instruments. Large cleanrooms typically contain multiple interconnected or isolated clean spaces, each requiring precise pressure differentials to prevent cross-contamination. Therefore, effectively controlling multiple FFU groups to achieve coordinated dynamic balance of pressure differentials across these areas is a core technology for ensuring a qualified production environment and stable product quality. Current technologies for controlling multi-area FFUs typically employ distributed independent control or simple linkage control strategies. In distributed control, each clean space's FFU controller only performs PID adjustments based on feedback from its own area's pressure differential sensor, lacking inter-area coordination. While simple linkage control considers the mutual influence between areas, it is mostly based on fixed logic rules or static models. For example, if the FFU speed in one area changes, the FFUs in adjacent areas will also adjust according to a preset ratio.

[0003] However, existing technical solutions have significant drawbacks. Due to the complexity and dynamism of the cleanroom environment, factors such as personnel movement, equipment door opening and closing, and changes in process exhaust ventilation can all have complex and nonlinear effects on the pressure differentials of multiple areas. Distributed control cannot handle the coupling effects between areas, often leading to fluctuations and continuous oscillations in pressure differentials. Simple linkage control based on fixed rules lacks adaptability to dynamic changes, cannot predictively handle environmental disturbances, resulting in lag in control response and low accuracy, making it difficult to optimize system energy consumption while ensuring stable pressure differentials in all areas. Summary of the Invention

[0004] To address the aforementioned issues, this invention provides a control method for the coordinated dynamic balance of pressure differentials in multiple regions of FFUs. This method employs a combination of predictive control and adaptive optimization, enabling coordinated stabilization of pressure differentials across multiple regions and optimization of energy consumption.

[0005] The above objectives can be achieved through the following approach:

[0006] A method for controlling the dynamic balance of differential pressure in multiple FFUs (Fan Filter Units) in multiple clean spaces includes: real-time acquisition of FFU operating parameters, environmental parameters, and energy consumption data in multiple clean spaces, and acquisition of corresponding historical data for each clean space, wherein the environmental parameters include differential pressure data, particle concentration data, and personnel flow data; based on the historical data, constructing an energy consumption assessment model to assess the energy consumption status of FFUs in each clean space in real time, comparing it with the real-time acquired energy consumption data, and optimizing the parameters of the energy consumption assessment model; inputting the real-time acquired FFU operating parameters and environmental parameters into the optimized energy consumption assessment model to obtain predicted energy consumption data for each clean space over a future period; and establishing a multi-zone FFU dynamic balance control system based on the FFU operating parameters, environmental parameters, and historical data. A multidimensional prediction model is used to generate predicted pressure difference and personnel flow trends for each cleanroom over a future period. Based on the predicted energy consumption data, pressure difference trends, and personnel flow trends, an FFU group collaborative control strategy is generated for multiple cleanrooms. This strategy guides the operating modes and rotation speeds of FFUs within the multiple cleanrooms. Based on this strategy, dynamic adjustment commands are issued to the FFUs in each cleanroom. After the dynamic adjustment commands are issued, FFU operating parameters, environmental parameters, and energy consumption data are continuously collected. The multidimensional prediction model and energy consumption assessment model are adaptively optimized based on the continuously collected parameters to ensure a continuous dynamic balance of pressure difference across multiple FFU zones.

[0007] Optionally, based on the historical data, an energy consumption assessment model is constructed to evaluate the energy consumption status of each FFU in the clean space in real time. This includes: using a reinforcement learning algorithm to initially train the energy consumption assessment model to be trained based on the FFU operating parameters, environmental parameters, and energy consumption data contained in the historical data, wherein the reinforcement learning algorithm uses the energy consumption data in the historical data as the reward function; inputting the FFU operating parameters and environmental parameters before the current moment into the energy consumption assessment model to obtain the current moment's energy consumption assessment result of the FFU; comparing the current moment's energy consumption assessment result with the real-time collected energy consumption data to obtain the assessment difference, and using an error backpropagation algorithm to optimize the parameters of the energy consumption assessment model based on the assessment difference.

[0008] Optionally, the step of optimizing the parameters of the energy consumption assessment model using the backpropagation algorithm based on the assessment difference includes: assigning dynamic weights to the assessment difference based on the assessment difference, pressure difference data, and traffic flow data; using the assessment difference with assigned dynamic weights, calculating the gradient value of each parameter in the energy consumption assessment model using the backpropagation algorithm, and updating the parameters of the energy consumption assessment model based on the gradient value to reduce the assessment difference.

[0009] Optionally, establishing the multidimensional prediction model includes: generating a time series sequence of FFU operating parameters and a time series sequence of environmental parameters from the historical data in chronological order; fusing the time series sequences of FFU operating parameters and environmental parameters with multimodal features to obtain a fused feature vector; and training the multidimensional prediction model to be trained using a deep neural network based on a long short-term memory network, with the fused feature vector as input and the predicted trends of pressure difference and traffic flow as output.

[0010] Optionally, generating an FFU group collaborative control strategy for multiple clean spaces based on the predicted energy consumption data, pressure difference change trend, and people flow change trend includes: pre-setting a group collaborative control strategy rule base containing FFU operating modes, speeds, and energy-saving control rules; assigning dynamic weights to each rule in the FFU group collaborative control strategy rule base using a fuzzy logic algorithm based on the predicted energy consumption data, pressure difference change trend, and people flow change trend; and matching and generating an FFU group collaborative control strategy from the FFU group collaborative control strategy rule base based on the dynamic weights, while simultaneously meeting the requirements of minimizing energy consumption and dynamically balancing pressure difference.

[0011] Optionally, the preset group collaborative control strategy rule base containing FFU operating modes, speed, and energy-saving control rules includes: an FFU operating mode rule set for FFU standby mode, low-speed operation mode, and full-speed operation mode; an FFU operating parameter rule set for FFU wind speed adjustment range and air volume adjustment range; and an FFU energy-saving control rule set for FFU differential pressure threshold, energy consumption threshold, and passenger flow threshold.

[0012] Optionally, based on the FFU group collaborative control strategy, issuing dynamic adjustment instructions to each FFU in the clean space includes: converting the FFU group collaborative control strategy into FFU dynamic adjustment instructions, wherein the FFU dynamic adjustment instructions include the FFU target operating mode and target rotation speed; issuing the FFU dynamic adjustment instructions to each FFU in the clean space, and during the issuance process, prioritizing the execution of instructions with higher energy consumption optimization weight and higher pressure differential balance weight.

[0013] Optionally, the step of prioritizing the execution of instructions with higher energy consumption optimization weights and higher differential pressure balance weights during the issuance process includes: assessing the degree of energy consumption anomaly of each FFU in the FFU group collaborative control strategy based on the real-time collected energy consumption data and predicted energy consumption data, and assigning an energy consumption optimization weight to each FFU according to the degree of energy consumption anomaly; assessing the contribution of each FFU in the clean space to differential pressure stability based on the real-time collected differential pressure data and differential pressure change trends, and assigning a differential pressure balance weight to each FFU according to the contribution; calculating the comprehensive priority score of the FFU dynamic adjustment instructions using a weighted summation algorithm based on the energy consumption optimization weights and differential pressure balance weights, and executing the dynamic adjustment instructions in descending order of comprehensive priority score.

[0014] Optionally, the adaptive optimization of the multidimensional prediction model and energy consumption assessment model based on continuously collected parameters to ensure the continuous dynamic balance of FFU pressure differentials in multiple regions includes: using the continuously collected FFU operating parameters, environmental parameters, and energy consumption data as incremental datasets to incrementally learn the multidimensional prediction model and energy consumption assessment model, and updating the model parameters; performing performance evaluation on the updated model; if the performance evaluation result is better than the current model, then the model update is completed; if the performance evaluation result is not better than the current model, then the model is restored to the state before the update.

[0015] Based on the same inventive concept, this invention also provides a control system for dynamic balance of differential pressure in multiple FFUs (Fan Filter Units) in multiple clean spaces. The system includes: a data acquisition module for real-time acquisition of FFU operating parameters, environmental parameters, and energy consumption data in multiple clean spaces, and for obtaining historical data; an energy consumption assessment module for constructing an energy consumption assessment model based on the historical data and for real-time assessment of the energy consumption status of FFUs in each clean space; a multi-dimensional prediction module for establishing a multi-dimensional prediction model based on the FFU operating parameters, environmental parameters, and historical data, and generating predicted differential pressure change trends and passenger flow change trends in each clean space over a future period; a collaborative control strategy module for generating a collaborative control strategy for multiple clean spaces based on the predicted energy consumption data, differential pressure change trends, and passenger flow change trends; an instruction issuance module for issuing dynamic adjustment instructions to FFUs in each clean space based on the FFU group collaborative control strategy; and an adaptive optimization module for continuously acquiring FFU operating parameters, environmental parameters, and energy consumption data after the dynamic adjustment instructions are issued, and adaptively optimizing the multi-dimensional prediction model and energy consumption assessment model based on the continuously acquired parameters.

[0016] Compared with the prior art, the present invention has the following advantages:

[0017] 1. This invention, by constructing a multidimensional prediction model, can proactively predict the changing trends of both the flow of people and the pressure differential itself, which affect the pressure differential in clean spaces. This predictive capability enables a shift from traditional delayed response to proactive pre-regulation, allowing intervention measures to be taken before or at the initial stage of environmental disturbances, thereby effectively suppressing pressure differential fluctuations and greatly improving the stability and reliability of multi-area clean environments.

[0018] 2. This invention integrates predicted energy consumption data, pressure difference change trends, and passenger flow change trends as a comprehensive decision-making basis for generating a collaborative control strategy. This multi-objective fusion decision-making mechanism ensures that the two core objectives of pressure difference stability and energy consumption economy can be intelligently balanced during control execution. Under the premise of ensuring cleanliness requirements, it dynamically finds the globally optimal operating scheme, realizing the synergistic optimization of pressure difference control and energy saving.

[0019] 3. This invention designs a complete closed-loop adaptive optimization process, which continuously collects operational data to incrementally learn and verify the energy consumption assessment model and the multidimensional prediction model. This endows the system with the ability to self-evolve and adapt to environmental changes, enabling it to dynamically track and adapt to long-term operating condition changes such as equipment aging and filter clogging, thereby ensuring the effectiveness and robustness of the control method during long-term operation and maintaining the continuous high efficiency of system performance.

[0020] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a flowchart illustrating a control method for coordinated dynamic balance of pressure differentials in multi-region FFUs according to an embodiment of the present invention.

[0023] Figure 2 This is a comparison chart of system performance under different control strategies according to embodiments of the present invention.

[0024] Figure 3 This is a timing diagram of the FFU dynamic adjustment instruction priority in an embodiment of the present invention.

[0025] Figure 4This is a schematic diagram of the structure of a multi-region FFU differential pressure collaborative dynamic balance control system according to an embodiment of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0027] Reference Figure 1 One embodiment of the present invention proposes a control method for coordinated dynamic balance of pressure differentials in multiple FFUs (Fan Filter Units). This method employs a combination of predictive control and adaptive optimization, achieving coordinated stability of pressure differentials and optimization of energy consumption across multiple zones, significantly improving the stability and operational efficiency of the multi-zone clean environment. Specifically, the method of this embodiment includes:

[0028] Real-time acquisition of FFU operating parameters, environmental parameters, and energy consumption data in multiple clean spaces, and acquisition of corresponding historical data in multiple clean spaces, wherein the environmental parameters include differential pressure data, particle concentration data, and personnel flow data;

[0029] Specifically, the system acquires real-time operating parameters such as airflow speed, rotation speed, and power from each FFU (Fan Filter Unit) within the cleanroom. Simultaneously, sensors distributed throughout the cleanroom sense and acquire real-time data on ambient pressure differential, particle concentration, and passenger flow. Furthermore, it obtains historical data on FFU operating parameters, environmental parameters, and energy consumption for each cleanroom, storing these data to form both a real-time data stream and a historical database.

[0030] Based on the historical data, an energy consumption assessment model is constructed to assess the energy consumption status of each FFU in the clean space in real time, and the model is compared with the real-time collected energy consumption data to optimize the parameters of the energy consumption assessment model.

[0031] Specifically, an energy consumption assessment model is constructed. Using historical FFU operating parameters and environmental parameters as input, the model calculates the expected energy consumption of the FFU and compares the expected results with the energy consumption data in the historical data to train the energy consumption assessment model. In addition, the FFU operating parameters and environmental parameters before the current moment are input into the trained energy consumption assessment model to obtain the expected energy consumption data at the current moment. The expected energy consumption data at the current moment is compared with the real-time collected energy consumption data to optimize the model parameters and improve the assessment accuracy.

[0032] The real-time collected FFU operating parameters and environmental parameters are input into the optimized energy consumption assessment model to obtain the predicted energy consumption data for each clean space in the future.

[0033] Specifically, real-time collected FFU operating parameters and environmental parameters are input into an optimized energy consumption assessment model. Based on these inputs, the energy consumption trend of the FFU can be predicted over the next hour or longer. For example, if the FFU's rotational speed is continuously increasing, the model can predict that its energy consumption will increase accordingly in the future.

[0034] Based on the FFU operating parameters, environmental parameters and historical data, a multidimensional prediction model is established to generate the predicted pressure difference change trend and the flow of people change trend in each clean space over a period of time.

[0035] Specifically, historical data is used to train a multidimensional prediction model. This model learns the complex nonlinear relationships between FFU operating status, environmental changes, pressure differential, and pedestrian traffic. In real-time operation, the model can predict potential future pressure differential and pedestrian traffic changes for each cleanroom based on the current state. For example, the model can predict the peak pedestrian traffic period for the next hour and subsequently predict the downward trend of the pressure differential.

[0036] Based on the predicted energy consumption data, pressure difference change trend, and personnel flow change trend, an FFU group collaborative control strategy is generated for multiple clean spaces. The FFU group collaborative control strategy is used to guide the working mode and rotation speed of FFUs in multiple clean spaces.

[0037] Specifically, after receiving prediction data from the energy consumption assessment model and the multi-dimensional prediction model, a group collaborative control strategy is generated for all FFUs in the clean space based on this prediction data. The generation process of this strategy comprehensively considers pressure differential stability and energy consumption economy, representing a multi-objective optimization decision. For example, when a decrease in pressure differential is predicted but energy consumption is high, a collaborative control strategy will be generated. This strategy may include instructions to reduce the rotational speed of some FFUs while increasing the rotational speed of others, in order to optimize energy consumption while maintaining overall pressure differential stability.

[0038] Based on the FFU group collaborative control strategy, dynamic adjustment instructions are issued to each FFU in the clean space;

[0039] Specifically, the generated FFU group collaborative control strategy is transformed into dynamic adjustment commands for the FFUs. These commands include the target operating mode and target speed for each FFU. These commands are then sent to the FFUs within the cleanroom via wired or wireless networks, and the FFUs immediately adjust their fans accordingly upon receiving the commands.

[0040] After the dynamic adjustment command is issued, FFU operating parameters, environmental parameters and energy consumption data are continuously collected, and the multidimensional prediction model and energy consumption assessment model are adaptively optimized based on the continuously collected parameters to ensure the continuous dynamic balance of FFU pressure difference in multiple regions.

[0041] Specifically, after the FFU executes the command, it continuously collects new data. This new data is used to incrementally learn the energy consumption assessment model and the multidimensional prediction model, continuously optimizing their prediction and assessment capabilities. This closed-loop feedback mechanism ensures that the control method can dynamically adapt to long-term changes such as equipment aging and filter clogging, thereby guaranteeing the continuous high efficiency and robustness of the system performance.

[0042] By employing a technical solution that combines real-time multidimensional data acquisition, energy consumption assessment models, and multidimensional prediction models, it is possible to accurately predict the changing trends of pressure differential and passenger flow while minimizing energy consumption. Furthermore, through an FFU group collaborative control strategy, it can proactively rather than passively adjust the operating status of FFUs in multiple areas, significantly improving the dynamic balance of pressure differential in clean spaces, system operating efficiency, and energy-saving effects.

[0043] Optionally, based on the historical data, an energy consumption assessment model is constructed to assess the energy consumption status of each FFU in the clean space in real time, including:

[0044] Based on the FFU operating parameters, environmental parameters and energy consumption data contained in the historical data, a reinforcement learning algorithm is used to initially train the energy consumption evaluation model to be trained, wherein the reinforcement learning algorithm uses the energy consumption data in the historical data as the reward function.

[0045] Specifically, in the initial training phase of the energy consumption assessment model, the operating parameters, environmental parameters, and energy consumption data of the FFU (Functional Functional Unit) over historical periods are used as the historical sample set. The model is trained using a reinforcement learning algorithm, aiming to learn an optimal policy that accurately predicts the FFU's energy consumption given the FFU's operating and environmental parameters. The historical energy consumption data is used as the reward function in reinforcement learning; the model continuously adjusts its parameters to maximize this reward, thereby improving the accuracy of energy consumption prediction. For example, a high reward is given when the model's predicted energy consumption is very close to the actual energy consumption, and a low reward is given if the predicted energy consumption is far from the actual energy consumption.

[0046] Input the FFU operating parameters and environmental parameters from the current moment to the energy consumption assessment model to obtain the current energy consumption assessment result of the FFU.

[0047] Specifically, after the model completes its initial training, the FFU operating parameters and environmental parameters up to the current time step are used as input features and fed into the energy consumption assessment model. The energy consumption assessment model immediately processes these inputs and outputs the FFU's energy consumption assessment result at the current time step. This achieves real-time, intelligent assessment of FFU energy consumption.

[0048] The energy consumption assessment result at the current moment is compared with the energy consumption data collected in real time to obtain the assessment difference. Based on the assessment difference, the parameters of the energy consumption assessment model are optimized using the error backpropagation algorithm.

[0049] Specifically, the energy consumption assessment result output by the energy consumption assessment model at the current moment is compared with the actual energy consumption data collected in real time, and the difference between the two assessments is calculated. Then, using an error backpropagation algorithm, this difference is used as an error signal and passed back to the energy consumption assessment model, and the model's internal parameters are adjusted based on this error signal. This continuous closed-loop optimization mechanism ensures that the energy consumption assessment model can adapt to changes in energy consumption characteristics caused by long-term factors such as FFU equipment aging and filter clogging, maintaining its assessment accuracy and robustness. Figure 2 As shown, the overall performance advantages of the present invention and traditional control in terms of differential pressure stability and energy consumption are intuitively compared.

[0050] Optionally, the step of optimizing the parameters of the energy consumption assessment model using an error backpropagation algorithm based on the assessment difference includes:

[0051] Based on the assessment difference, pressure difference data, and pedestrian flow data, dynamic weights are assigned to the assessment difference;

[0052] Specifically, after obtaining the difference between the FFU energy consumption assessment result and the actual energy consumption data, a dynamic weight needs to be assigned to this difference. The weight allocation is based on the current differential pressure data and personnel flow data. For example, when the differential pressure in the clean space deviates significantly from the set value or when personnel flow is at its peak, it indicates that the control environment is in an unstable or high-load state. In this case, the assessment difference is more important for model optimization, and a larger dynamic weight is assigned. This ensures that the model can learn and adapt to energy consumption changes in high-priority environments more preferentially.

[0053] Based on the evaluation difference with dynamically assigned weights, the backpropagation algorithm is used to calculate the gradient value of each parameter in the energy consumption evaluation model, and the parameters of the energy consumption evaluation model are updated based on the gradient value to reduce the evaluation difference.

[0054] Specifically, the evaluation difference, assigned with dynamic weights, is used as the output of the loss function, and the gradient value of each parameter in the energy consumption assessment model is calculated using the error backpropagation algorithm. Based on these gradient values, optimization algorithms such as gradient descent are used to update the model's parameters. Through this process, the model can learn in a targeted manner according to the dynamic weights, more effectively reducing the evaluation difference, improving the accuracy of energy consumption assessment, and enhancing the model's adaptability to complex operating conditions.

[0055] Optionally, establishing the multidimensional prediction model includes:

[0056] The FFU operating parameters and environmental parameters contained in the historical data are used to generate FFU operating parameter time series and environmental parameter time series respectively in chronological order;

[0057] Specifically, the timing sequence of FFU operating parameters can be represented as follows:

[0058] ,

[0059] in, At any moment A vector of collected FFU parameters such as wind speed, rotational speed, and power. The time series sequence of environmental parameters can be represented as...

[0060] ,

[0061] in, At any moment Vectors of collected parameters such as pressure difference, particle concentration, and pedestrian flow are generated. These data vectors are arranged in chronological order to form their respective time series. To ensure the accuracy of the model, these time series have undergone data cleaning and preprocessing, such as missing value imputation, outlier removal, and normalization, to eliminate the interference of different units on the calculation and ensure the data has mathematical validity.

[0062] The time series sequences of FFU operating parameters and environmental parameters are fused using multimodal features to obtain a fused feature vector;

[0063] Specifically, to capture the interaction between the FFU's operating state and the environmental state, a multimodal feature fusion technique is employed. The fusion process is represented by the following formula:

[0064] ,

[0065] in, At any moment The resulting fused feature vector, , , It is a learnable fusion coefficient, the specific value of which is determined through model training. "" represents element-wise multiplication. This fusion process combines weighted summation with interaction terms to integrate data from different modalities into a unified vector, which can more comprehensively describe the dynamic state of the clean space and capture the nonlinear correlation between different parameters.

[0066] A deep neural network based on long short-term memory is used, with the fused feature vector as input, and the predicted pressure difference change trend and traffic flow change trend as output, to train the multidimensional prediction model to be trained.

[0067] Specifically, a deep neural network based on Long Short-Term Memory (LSTM) is used as the multidimensional prediction model. LSTM networks are particularly adept at processing and predicting time-series data, capable of capturing long-term dependencies in the data, such as the impact of historical pedestrian flow fluctuations on future pressure differential trends. The training process can be represented as minimizing the loss function:

[0068] ,

[0069] in, and They represent the times at time 1 and 2 respectively. The model's predicted pressure difference and the actual pressure difference, and They represent the times at time 1 and 2 respectively. The model's predicted pedestrian flow and actual pedestrian flow. This represents the total number of historical data points. By minimizing this loss function, the trained model can accurately predict future trends in pressure differential and pedestrian traffic based on the fused feature vectors. This training process can employ optimization algorithms such as gradient descent and use cross-validation to adjust hyperparameters, ensuring the model's generalization ability and prediction accuracy under different operating conditions.

[0070] Optionally, based on the predicted energy consumption data, pressure difference change trends, and people flow change trends, an FFU group collaborative control strategy is generated for multiple clean spaces, including:

[0071] A pre-defined group collaborative control strategy rule base includes FFU operating modes, speed, and energy-saving control rules;

[0072] Specifically, before generating the FFU group collaborative control strategy, a rule base containing various control rules needs to be preset. This rule base can include three categories of rules: FFU operating mode rules, FFU operating parameter rules, and FFU energy-saving control rules. For example, operating mode rules can define "standby mode," "low-speed operation mode," and "full-speed operation mode," etc.; operating parameter rules can define the FFU's fan speed adjustment range and airflow adjustment range; energy-saving control rules can define the FFU's control strategy under different pressure differentials, energy consumption, and passenger flow thresholds. The combination of these rules constitutes the decision space for group collaborative control.

[0073] Based on the predicted energy consumption data, pressure difference change trend and population flow change trend, a fuzzy logic algorithm is used to assign dynamic weights to each rule in the FFU group collaborative control strategy rule base.

[0074] Specifically, fuzzy logic algorithms are used to process fuzzy and uncertain information such as predicted energy consumption data, pressure difference trends, and pedestrian flow trends. First, these predicted data are used as input variables for the fuzzy algorithm; for example, "predicted energy consumption" can be defined as a fuzzy set of "low," "medium," or "high." "Pressure difference trends" and "pedestrian flow trends" undergo similar fuzzification. Next, the fuzzy algorithm assigns a dynamic weight to each rule in the rule base according to preset fuzzy rules. For instance, when predicted energy consumption is "high" and the pressure difference trend is "decreasing," the rule related to "reducing FFU speed" will receive a higher weight. This can be represented as a fuzzy inference function:

[0075] ,

[0076] in, These are the dynamic weights assigned to the rules. It is predicted energy consumption data. It is the trend of pressure difference change. It is the trend of changing pedestrian traffic. It is a fuzzy logic inference function. Through this process, fuzzy logic algorithms can transform complex prediction information into clear, quantifiable decision weights.

[0077] Based on the dynamic weights, an FFU group collaborative control strategy is matched and generated from the FFU group collaborative control strategy rule base, which simultaneously meets the requirements of minimizing energy consumption and dynamic pressure balance.

[0078] Specifically, after assigning dynamic weights to each rule in the rule base, a final FFU group collaborative control strategy is generated based on the weights. A weighted average or maximum membership method can be used to select the rule with the highest weight or a combination of multiple rules from the rule base as the final control strategy. This strategy aims to simultaneously satisfy the dual objectives of minimizing energy consumption and maintaining dynamic pressure balance. For example, the generated strategy might instruct some high-energy-consuming FFUs to enter a low-speed mode, while instructing other FFUs that contribute significantly to pressure stability to maintain their current speed, thereby achieving collaborative optimization of energy consumption and pressure difference throughout the cleanroom.

[0079] Optionally, the preset group collaborative control strategy rule base, which includes FFU operating modes, speed, and energy-saving control rules, includes:

[0080] FFU operating mode rule set for FFU standby mode, low-speed operation mode and full-speed operation mode;

[0081] Specifically, the FFU operating mode rule set defines the basic operating modes of the FFU under different conditions. The standby mode rule refers to the FFU entering a dormant or low-power state when the cleanroom is unoccupied, energy consumption is extremely low, and the differential pressure is maintained above the safe threshold. The low-speed operation mode rule refers to the FFU operating at a lower speed when energy consumption optimization is the primary focus and the differential pressure change trend is gradual. The full-speed operation mode rule refers to the FFU operating at its highest speed in emergency situations such as a rapid drop in differential pressure, a sharp increase in particle concentration, or a sudden increase in personnel flow, in order to quickly restore differential pressure and cleanliness.

[0082] FFU operating parameter rule set for FFU wind speed adjustment range and air volume adjustment range;

[0083] Specifically, the FFU operating parameter rule set specifies the adjustment range of the FFU in different modes. The wind speed adjustment range rule defines the permissible wind speed range for the FFU in low, medium, and high speed modes. For example, in low speed mode, the adjustable wind speed range is... arrive The airflow adjustment range rules define the upper and lower limits of airflow adjustment, ensuring that the FFU can provide the minimum airflow required to meet cleanliness requirements during operation, while not exceeding its designed maximum airflow capacity. The airflow calculation formula is as follows:

[0084] ,

[0085] in, For air volume, The cross-sectional area of ​​the air outlet. This refers to wind speed.

[0086] FFU energy-saving control rule set including FFU differential pressure threshold, energy consumption threshold and people flow threshold.

[0087] Specifically, the FFU energy-saving control rule set defines the threshold conditions for triggering FFU mode switching and speed adjustment. The differential pressure threshold rule specifies the minimum differential pressure value required to maintain positive pressure in the cleanroom. and maximum pressure difference When the pressure difference exceeds this range, adjustments are required. The energy consumption threshold rule sets the upper limit of the expected energy consumption of the FFU, such as... When actual energy consumption exceeds this threshold, the control strategy will prioritize energy-saving measures. The people flow threshold rule defines the cleanliness requirements under different people flow conditions; for example, when the people flow exceeds... At times, it will switch to a more aggressive control mode to address the risk of increased pollution.

[0088] Optionally, based on the FFU group collaborative control strategy, issuing dynamic adjustment instructions to each FFU in the clean space includes:

[0089] The FFU group collaborative control strategy is transformed into FFU dynamic adjustment instructions, wherein the FFU dynamic adjustment instructions include the FFU target operating mode and target speed;

[0090] Specifically, after generating the FFU group coordinated control strategy, it needs to be parsed into executable dynamic adjustment instructions. This process transforms abstract control strategies, such as "saving energy in area A and maintaining stable differential pressure in area B," into specific, identifiable FFU operation commands. For example, the instruction might require the FFUs in area A to enter "low-speed operation mode" and set the speed to 1000 rpm, while requiring the FFUs in area B to enter "full-speed operation mode" and set the speed to 2000 rpm.

[0091] The FFU dynamic adjustment command is issued to each FFU in the clean space, and during the issuance process, the command with higher energy consumption optimization weight and higher pressure differential balance weight is executed first.

[0092] Specifically, when issuing dynamic adjustment commands to FFUs in multiple cleanrooms, the commands are prioritized based on their respective priorities. The priority of a command is determined by both its energy consumption optimization weight and its pressure differential balance weight. For example, a command to correct energy consumption anomalies or a command to restore pressure differential in the core area will be assigned a higher priority. This prioritization mechanism ensures that, under network congestion or concurrent command conditions, commands most critical to the overall cleanroom stability and energy consumption optimization are executed by the FFUs first, thereby guaranteeing response speed and reliability. Figure 3 As shown, this illustrates the instruction priority scheduling mechanism.

[0093] Optionally, the step of prioritizing the execution of instructions with higher energy consumption optimization weight and higher pressure differential balance weight during the issuance process includes:

[0094] Based on the real-time collected energy consumption data and predicted energy consumption data, the degree of energy consumption anomaly of each FFU in the FFU group collaborative control strategy is evaluated, and energy consumption optimization weights are assigned to each FFU according to the degree of energy consumption anomaly.

[0095] Specifically, the allocation of energy consumption optimization weights is based on the degree of energy consumption anomaly of the FFU. The formula for calculating the degree of energy consumption anomaly is:

[0096] ,

[0097] in, It refers to the degree of abnormal energy consumption of the FFU. It is real-time collected energy consumption data. This is energy consumption data predicted by the energy consumption assessment model. When the energy consumption anomaly of an FFU exceeds a preset threshold, the energy consumption optimization weight of that FFU will be increased accordingly. This ensures that when energy consumption anomalies occur, control commands for these highly abnormal FFUs can be prioritized, achieving the goal of rapid energy saving.

[0098] Based on the real-time collected differential pressure data and differential pressure change trends, the contribution of each FFU in the clean space to differential pressure stability is evaluated, and a differential pressure balance weight is assigned to each FFU according to the contribution.

[0099] Specifically, the allocation of differential pressure balance weights is based on the contribution of the FFU (Fan Filter Unit) to differential pressure stability. The formula for evaluating the contribution is:

[0100] ,

[0101] in, It is the contribution of the FFU to differential pressure stability. It is real-time collected differential pressure data. This represents the slope of the pressure differential change trend, indicating the severity of the pressure differential change. When the pressure differential in the clean space where the FFU is located changes drastically and the adjustment of the FFU can significantly improve this change, the pressure differential balance weight of that FFU will increase accordingly. This ensures that when pressure differential fluctuates, the FFU that contributes the most to pressure differential stability will receive priority in executing commands to quickly restore pressure differential balance.

[0102] Based on the energy consumption optimization weight and pressure difference balance weight, a weighted summation algorithm is used to calculate the comprehensive priority score of the FFU dynamic adjustment command, and the dynamic adjustment command is executed in descending order according to the comprehensive priority score.

[0103] Specifically, the formula for the overall priority score of the FFU dynamic adjustment instruction is as follows:

[0104] ,

[0105] in, It is a comprehensive priority score. and These are preset global weighting coefficients used to adjust the relative importance of energy consumption optimization and pressure balance in decision-making. For example, in operating conditions where energy consumption optimization is the primary focus, It can be set to higher than After obtaining the overall priority score, dynamic adjustment instructions will be issued to the corresponding FFUs in descending order of score to ensure that critical instructions can be executed first.

[0106] Optionally, the adaptive optimization of the multidimensional prediction model and energy consumption assessment model based on continuously collected parameters to ensure the continuous dynamic balance of FFU pressure differentials in multiple regions includes:

[0107] The continuously collected FFU operating parameters, environmental parameters, and energy consumption data are used as incremental datasets to incrementally learn the multidimensional prediction model and energy consumption assessment model, and update the model parameters.

[0108] Specifically, after the FFU executes the command, it continuously collects new data. This new data forms an incremental dataset, which is used to incrementally learn the multidimensional prediction model and the energy consumption assessment model. Unlike traditional batch learning, incremental learning does not require retraining the entire model. Instead, it uses the new dataset to fine-tune the model parameters, enabling it to dynamically adapt to long-term operating condition changes such as equipment aging and filter clogging. For example, when filter clogging of the FFU leads to increased air resistance, its operating parameters will change. Through incremental learning, the energy consumption assessment model can learn this new energy consumption pattern and update its parameters accordingly.

[0109] The updated model is evaluated for performance. If the performance evaluation result is better than the current model, the model update is completed; if the performance evaluation result is not better than the current model, the model is restored to its state before the update.

[0110] Specifically, after the model parameters are updated, the updated model is immediately evaluated for performance. The performance of the new model is judged by metrics such as error rate and accuracy on the validation dataset. If the evaluation results show that the updated model outperforms the currently used model, the new model will be officially deployed. Conversely, if the new model performs poorly, to prevent an unstable model from affecting control accuracy, the model will be rolled back to its pre-update state. This adaptive optimization process with performance validation and rollback mechanisms ensures the continuous stability and reliability of the system.

[0111] Based on the same inventive concept, such as Figure 4 As shown, the present invention also provides a control system for coordinated dynamic balancing of multi-region FFU differential pressure, the system comprising:

[0112] The data acquisition module is used to collect FFU operating parameters, environmental parameters, and energy consumption data in multiple clean spaces in real time, and to acquire historical data;

[0113] The energy consumption assessment module is used to construct an energy consumption assessment model based on the historical data and to assess the energy consumption status of each FFU in the clean space in real time.

[0114] The multidimensional prediction module is used to establish a multidimensional prediction model based on the FFU operating parameters, environmental parameters and historical data, and generate the predicted pressure difference change trend and personnel flow change trend in each clean space over a future period of time.

[0115] The collaborative control strategy module is used to generate FFU group collaborative control strategies for multiple clean spaces based on the predicted energy consumption data, pressure difference change trend and people flow change trend.

[0116] The instruction issuing module is used to issue dynamic adjustment instructions to each FFU in the clean space based on the FFU group collaborative control strategy;

[0117] The adaptive optimization module is used to continuously collect FFU operating parameters, environmental parameters and energy consumption data after the dynamic adjustment command is issued, and to adaptively optimize the multidimensional prediction model and energy consumption assessment model based on the continuously collected parameters.

[0118] Example 1:

[0119] To verify the feasibility of this invention, it was applied to the cleanroom environment of a large semiconductor manufacturing plant. This plant has multiple interconnected clean areas with extremely stringent requirements for environmental parameters such as internal pressure differential and particle concentration. Traditional FFU control methods struggle to balance stable pressure differential with high operating energy consumption. This plant aims to use the method of this invention to achieve intelligent collaborative control of multi-area FFU groups, achieving the dual goals of dynamic pressure differential balance and optimal operating energy consumption.

[0120] In this embodiment, the semiconductor factory deploys the control method of this invention in its three core production areas: Area A, Area B, and Area C. The control method collects operating parameters, environmental parameters, and energy consumption data of the FFUs (Functional Fuses) through sensors deployed in each area. Using historical data from the past six months, a gradient boosting decision tree algorithm is employed for offline initial training of the energy consumption assessment model. Simultaneously, a multi-dimensional prediction model is constructed using a long short-term memory network to predict the pressure difference and pedestrian flow trends within the next 15 minutes.

[0121] During continuous operation testing in April, its dynamic optimization and forward-looking control capabilities were demonstrated. At 9:00 AM on April 15th, coinciding with shift change, the multi-dimensional prediction model, based on historical patterns, successfully predicted a surge in pedestrian traffic in Zone C from 5 to 25 people within the next 10 minutes, and also predicted a rapid decrease in pressure differential of -2.5 Pa. Simultaneously, the energy consumption assessment model, combined with real-time FFU parameters, output a real-time energy consumption assessment result of 4.5 kW for the FFUs in Zone C, while the real-time energy consumption collected by the power monitoring equipment was 4.7 kW, resulting in a 0.2 kW energy consumption difference. Because both pedestrian traffic and pressure differential trends pointed to high risk, the FFU commands were assigned a high dynamic weight, with a weight of [weight missing]. This accelerates the online parameter fine-tuning of the energy consumption assessment model, making it more accurately reflect the current operating conditions.

[0122] Based on the aforementioned predictive information, the fuzzy logic decision-making algorithm assigned dynamic weights to the rules in the collaborative control strategy rule base. At this point, the rule for "full-speed operation mode," which aims to "rapidly increase wind speed to stabilize pressure differential," received the highest weight. A collaborative control strategy was then generated and transformed into specific FFU dynamic adjustment commands. A command for zone C, specifying "target operating mode: full-speed operation, target speed: 1800 RPM," received a comprehensive priority score of 0.95 due to its highest contribution to maintaining pressure differential stability and was issued first. Meanwhile, zones A and B were stable, and their FFU energy-saving adjustment commands had priority scores of only 0.2-0.3, therefore execution was temporarily suspended. After the commands were issued, the pressure differential in zone C fluctuated by less than ±0.5 Pa during peak periods, far exceeding the ±3.0 Pa fluctuation range under traditional control, effectively avoiding the risk of cross-contamination caused by pressure differential imbalance.

[0123] During a month-long test, this invention reduced overall energy consumption by approximately 18% compared to the control group, while ensuring that the differential pressure compliance rate in all areas remained above 99.8%. It also achieved a leakage prediction accuracy of over 92%, and through optimized scheduling, reduced the average response and intervention time for high-risk events from 2 minutes in traditional control to less than 10 seconds.

[0124] Table 1. Prediction Accuracy Data of Multidimensional Prediction Model

[0125]

[0126] Table 2 Comparison of System Performance under Different Control Strategies

[0127]

[0128] Table 3 Examples of FFU Dynamic Adjustment Command Priority

[0129]

[0130] Tables 1 to 3 above show the actual application data of the present invention in the cleanroom of a semiconductor factory, demonstrating in detail its superior performance in terms of prediction accuracy, collaborative control effect and intelligent scheduling.

[0131] The data in Table 1 show that the multidimensional prediction model has a high accuracy rate in predicting the trends of pedestrian flow and pressure difference. For example, during peak pedestrian flow periods in Zone C, the prediction accuracy rate reaches over 90%, providing a reliable basis for taking proactive intervention measures.

[0132] The data in Table 2 clearly demonstrate the significant energy-saving advantages of this invention. After adopting this invention, the average pressure difference fluctuation was significantly reduced from ±3.2Pa to ±0.8Pa, the pass rate was increased to 99.8%, and an energy-saving effect of over 18% was achieved, successfully resolving the contradiction between pressure difference stability and energy saving.

[0133] Table 3 visually illustrates the operation of the instruction priority scheduling mechanism. When faced with the urgent risk of pressure imbalance in Zone C, the corresponding instruction is assigned a priority score as high as 0.95 and executed immediately, while non-urgent energy-saving instructions from other zones are placed in a waiting queue. This ensures that control resources are prioritized for handling the most critical issues, greatly improving response speed and robustness.

[0134] It should be noted that the electrical connections between the various units described above do not necessarily represent direct or indirect connections. Any indirect connection method is applicable to the embodiments of the present invention as long as it achieves the purpose of this invention. The above descriptions are merely exemplary embodiments of the present invention and should not be construed as limiting the scope of the invention.

[0135] All equivalent changes and modifications made in accordance with the teachings of this invention are still within the scope of this invention. Those skilled in the art will readily conceive of other embodiments of this invention upon considering the specification and the disclosure of practical truth. This application is intended to cover any variations, uses, or adaptations of this invention that follow the general principles of this invention and include common knowledge or conventional techniques in the art not described herein.

Claims

1. A control method for coordinated dynamic balance of pressure differentials in multi-region FFUs, characterized in that, The method includes: Real-time acquisition of FFU operating parameters, environmental parameters, and energy consumption data in multiple clean spaces, and acquisition of corresponding historical data in multiple clean spaces, wherein the environmental parameters include differential pressure data, particle concentration data, and personnel flow data; Based on the historical data, an energy consumption assessment model is constructed to evaluate the energy consumption status of each FFU in the clean space in real time. This evaluation is then compared with the real-time collected energy consumption data to optimize the parameters of the energy consumption assessment model. The process of constructing the energy consumption assessment model based on the historical data and evaluating the energy consumption status of each FFU in the clean space in real time includes: using a reinforcement learning algorithm to initially train the energy consumption assessment model based on the FFU operating parameters, environmental parameters, and energy consumption data contained in the historical data, where the reinforcement learning algorithm uses the energy consumption data from the historical data as the reward function; inputting the FFU operating parameters and environmental parameters from the previous time into the energy consumption assessment model to obtain the current energy consumption assessment result of the FFU; comparing the current energy consumption assessment result with the real-time collected energy consumption data to obtain the assessment difference; and using an error backpropagation algorithm based on the assessment difference to optimize the parameters of the energy consumption assessment model. The real-time collected FFU operating parameters and environmental parameters are input into the optimized energy consumption assessment model to obtain the predicted energy consumption data for each clean space in the future. Based on the FFU operating parameters, environmental parameters, and historical data, a multidimensional prediction model is established to generate predicted pressure difference and passenger flow trends for each cleanroom over a future period. The establishment of the multidimensional prediction model includes: generating time-series sequences of FFU operating parameters and environmental parameters from the historical data in chronological order; fusing multimodal features of the FFU operating parameter time-series sequences and the environmental parameter time-series sequences to obtain a fused feature vector; and training the multidimensional prediction model using a deep neural network based on a long short-term memory network, with the fused feature vector as input and the predicted pressure difference and passenger flow trends as output. Based on the predicted energy consumption data, pressure difference change trend, and personnel flow change trend, an FFU group collaborative control strategy is generated for multiple clean spaces. The FFU group collaborative control strategy is used to guide the working mode and rotation speed of FFUs in multiple clean spaces. Based on the FFU group collaborative control strategy, dynamic adjustment instructions are issued to each FFU in the clean space; After the dynamic adjustment command is issued, FFU operating parameters, environmental parameters and energy consumption data are continuously collected, and the multidimensional prediction model and energy consumption assessment model are adaptively optimized based on the continuously collected parameters to ensure the continuous dynamic balance of FFU pressure difference in multiple regions.

2. The control method for coordinated dynamic balance of multi-region FFU pressure differential as described in claim 1, characterized in that, The optimization of the energy consumption assessment model parameters using the backpropagation algorithm based on the assessment difference includes: Based on the assessment difference, pressure difference data, and pedestrian flow data, dynamic weights are assigned to the assessment difference; Based on the evaluation difference with dynamically assigned weights, the backpropagation algorithm is used to calculate the gradient value of each parameter in the energy consumption evaluation model, and the parameters of the energy consumption evaluation model are updated based on the gradient value to reduce the evaluation difference.

3. The control method for coordinated dynamic balance of multi-region FFU pressure differential as described in claim 1, characterized in that, Based on the predicted energy consumption data, pressure difference change trends, and personnel flow change trends, a collaborative control strategy for FFU groups is generated for multiple clean spaces, including: A pre-defined group collaborative control strategy rule base includes FFU operating modes, speed, and energy-saving control rules; Based on the predicted energy consumption data, pressure difference change trend and traffic flow change trend, a fuzzy logic algorithm is used to assign dynamic weights to each rule in the FFU group collaborative control strategy rule base. Based on the dynamic weights, an FFU group collaborative control strategy is matched and generated from the FFU group collaborative control strategy rule base, which simultaneously meets the requirements of minimizing energy consumption and dynamic pressure balance.

4. The control method for coordinated dynamic balance of multi-region FFU pressure differential as described in claim 3, characterized in that, The preset group collaborative control strategy rule base, which includes FFU operating modes, speed, and energy-saving control rules, includes: FFU operating mode rule set for FFU standby mode, low-speed operation mode and full-speed operation mode; FFU operating parameter rule set for FFU wind speed adjustment range and air volume adjustment range; FFU energy-saving control rule set based on FFU differential pressure threshold, energy consumption threshold, and passenger flow threshold.

5. The control method for coordinated dynamic balance of multi-region FFU pressure differential as described in claim 1, characterized in that, Based on the aforementioned FFU group collaborative control strategy, dynamic adjustment instructions are issued to each FFU in the clean space, including: The FFU group collaborative control strategy is transformed into FFU dynamic adjustment instructions, wherein the FFU dynamic adjustment instructions include the FFU target operating mode and target speed; The FFU dynamic adjustment command is issued to each FFU in the clean space, and during the issuance process, the command with higher energy consumption optimization weight and higher pressure differential balance weight is executed first.

6. The control method for coordinated dynamic balance of multi-region FFU pressure differential as described in claim 5, characterized in that, During the distribution process, instructions with higher energy consumption optimization weight and higher pressure differential balance weight are prioritized for execution, including: Based on the real-time collected energy consumption data and predicted energy consumption data, the degree of energy consumption anomaly of each FFU in the FFU group collaborative control strategy is evaluated, and energy consumption optimization weights are assigned to each FFU according to the degree of energy consumption anomaly. Based on the real-time collected differential pressure data and differential pressure change trends, the contribution of each FFU in the clean space to differential pressure stability is evaluated, and a differential pressure balance weight is assigned to each FFU according to the contribution. Based on the energy consumption optimization weight and pressure difference balance weight, a weighted summation algorithm is used to calculate the comprehensive priority score of the FFU dynamic adjustment command, and the dynamic adjustment command is executed in descending order according to the comprehensive priority score.

7. The control method for coordinated dynamic balance of multi-region FFU pressure differential as described in claim 1, characterized in that, The adaptive optimization of the multidimensional prediction model and energy consumption assessment model based on continuously collected parameters to ensure the continuous dynamic balance of FFU pressure differentials in multiple regions includes: The continuously collected FFU operating parameters, environmental parameters, and energy consumption data are used as incremental datasets to incrementally learn the multidimensional prediction model and energy consumption assessment model, and update the model parameters. The updated model is evaluated for performance. If the performance evaluation result is better than the current model, the model update is completed; if the performance evaluation result is not better than the current model, the model is restored to its state before the update.

8. A control system for coordinated dynamic balancing of multi-region FFU differential pressure is applied to the control method for coordinated dynamic balancing of multi-region FFU differential pressure as described in any one of claims 1-7, characterized in that, The system includes: The data acquisition module is used to collect FFU operating parameters, environmental parameters, and energy consumption data in multiple clean spaces in real time, and to acquire historical data; The energy consumption assessment module is used to construct an energy consumption assessment model based on the historical data and to assess the energy consumption status of each FFU in the clean space in real time. The process of constructing the energy consumption assessment model based on the historical data and assessing the energy consumption status of each FFU in the clean space in real time includes: initially training the energy consumption assessment model to be trained using a reinforcement learning algorithm based on the FFU operating parameters, environmental parameters, and energy consumption data contained in the historical data, wherein the reinforcement learning algorithm uses the energy consumption data in the historical data as the reward function; inputting the FFU operating parameters and environmental parameters from the previous time into the energy consumption assessment model to obtain the current energy consumption assessment result of the FFU; comparing the current energy consumption assessment result with the real-time collected energy consumption data to obtain the assessment difference; and optimizing the parameters of the energy consumption assessment model using an error backpropagation algorithm based on the assessment difference. A multidimensional prediction module is used to establish a multidimensional prediction model based on the FFU operating parameters, environmental parameters, and historical data, generating predicted pressure difference change trends and passenger flow change trends for each clean space over a future period. The establishment of the multidimensional prediction model includes: generating time-series sequences of FFU operating parameters and environmental parameters from the historical data in chronological order; fusing multimodal features of the FFU operating parameter time-series sequences and the environmental parameter time-series sequences to obtain a fused feature vector; and training the multidimensional prediction model using a deep neural network based on a long short-term memory network, with the fused feature vector as input and the predicted pressure difference change trends and passenger flow change trends as output. The collaborative control strategy module is used to generate FFU group collaborative control strategies for multiple clean spaces based on the predicted energy consumption data, pressure difference change trend and people flow change trend. The instruction issuing module is used to issue dynamic adjustment instructions to each FFU in the clean space based on the FFU group collaborative control strategy; The adaptive optimization module is used to continuously collect FFU operating parameters, environmental parameters and energy consumption data after the dynamic adjustment command is issued, and to adaptively optimize the multidimensional prediction model and energy consumption assessment model based on the continuously collected parameters.

Images