Variable air volume fume hood energy consumption control method and system based on surface air speed optimization

Abstract

The invention provides a variable air volume fume hood energy consumption control method and system based on surface air speed optimization, and relates to the technical field of air conditioning. Acquiring a gas concentration monitoring data set and an energy consumption monitoring data set of the variable air volume ventilation cabinet; respectively inputting the gas concentration monitoring data set and the energy consumption monitoring data set of the variable air volume fume hood into a preset harmful gas concentration dynamic prediction model and an energy consumption prediction model, and outputting a harmful gas concentration dynamic prediction value and an energy consumption prediction value; and finally, taking the minimum energy consumption as a target, considering a harmful gas concentration safety threshold constraint, solving the target based on a genetic algorithm, obtaining an optimal surface air speed sequence of the variable air volume fume hood, and solving a fume hood surface air speed control sequence with the lowest energy consumption on the premise of strictly ensuring that the harmful gas concentration is always below a safety threshold in the process. Therefore, collaborative optimization of operation safety and energy efficiency of the laboratory fume hood is realized. According to the invention, the energy-saving performance is obviously improved while the safety is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of air conditioning, and more particularly to a variable air volume fume hood energy consumption control method and system based on face velocity optimization. Background Art

[0002] Fume hoods are critical equipment used in the fields of chemistry, biology, and medicine to control hazardous gases, dust, or vapor. Their core function is to remove harmful gases, odors, dust, and other pollutants from the laboratory through directional airflow, protecting the safety of laboratory personnel and the health of the environment. The face velocity of a variable air volume fume hood refers to the air flow rate through the fume hood's operating port (i.e., the front glass window opening) and is a key indicator of fume hood performance and safety. Its primary function is to ensure the effective capture and discharge of harmful gases by controlling the dynamic balance of airflow, preventing pollutants from escaping into the laboratory environment.

[0003] In the early stages, fume hoods used a fixed air volume design, meaning that regardless of whether the fume hood windows were open or closed, the fan always ran at a constant speed to expel polluted gases out of the laboratory. However, this had the disadvantages of high energy consumption and difficulty in stabilizing the laboratory pressure differential. With the maturity of sensor technology and variable frequency speed regulation technology, variable air volume technology has gradually become popular in fume hood design. By monitoring parameters such as window height, pollutant concentration, and supply air speed, the supply and exhaust air volume is dynamically adjusted based on the PID algorithm to match the air volume with demand, improving the stability of the laboratory pressure differential. However, the PID control algorithm used in existing variable air volume systems is prone to air volume overshoot or oscillation in complex scenarios such as frequent window opening and closing or multi-parameter coupling, resulting in increased system energy consumption. In addition, air volume adjustment relies on real-time sensor data feedback from differential pressure sensors, temperature and humidity sensors, etc., but sensor signal delays or noise may cause the system to respond lags, especially in the event of sudden pollution.

[0004] The prior art proposes a variable air volume control method for a fume hood based on fuzzy control, which uses a fuzzy control algorithm to calculate the air volume adjustment value corresponding to the current laboratory according to the current environmental parameters; determines the air volume adjustment parameter of the fume hood according to the air volume adjustment value of the laboratory and the current air volume of the fume hood; and adjusts the air volume of the fume hood based on the air volume adjustment parameter of the fume hood, thereby achieving adaptive adjustment of the air volume of the fume hood in combination with the air state in the laboratory and improving the accuracy of air volume control. The fuzzy control algorithm handles the uncertainty of environmental parameters and combines with the PID controller to further optimize the response speed and steady-state error, ensuring accurate air volume adjustment and reducing overshoot or oscillation. However, the fuzzy control rules used require a lot of time to be jointly debugged with the parameters of the PID controller, and the time cost of deployment is high. At the same time, traditional fuzzy control rules and subordinate functions are fixed, and the fixed rule base is difficult to cover all scenarios and lacks the ability to respond dynamically. Summary of the Invention

[0005] In order to solve the problem that the method of fume hood air volume control cannot take into account the dual needs of safety and energy saving, the present invention proposes a variable air volume fume hood energy consumption control method and system based on surface wind speed optimization. While ensuring that the ventilation volume surface wind speed meets the safety requirements, it reduces the energy consumption during the use of the fume hood and meets the energy saving requirements.

[0006] In order to achieve the above technical effects, the technical solutions of the present invention are as follows:

[0007] In a first aspect, the present application proposes a variable air volume fume hood energy consumption control method based on face velocity optimization, comprising the following steps:

[0008] Obtain gas concentration monitoring data sets and energy consumption monitoring data sets for variable air volume fume hoods;

[0009] Input the gas concentration monitoring data set of the variable air volume fume hood into the preset dynamic prediction model of harmful gas concentration, and output the predicted value of harmful gas concentration; input the energy consumption monitoring data set into the preset energy consumption prediction model, and output the energy consumption prediction value;

[0010] With the goal of minimizing energy consumption and considering the safety threshold constraint of harmful gas concentration, the genetic algorithm is used to solve the goal and obtain the optimal face wind speed sequence of variable air volume fume hoods. Among them, the face wind speed sequence individuals of the variable air volume fume hoods are used as the population of the genetic algorithm, and the fitness value of the population is calculated based on the predicted value of harmful gas concentration and the predicted value of energy consumption.

[0011] In this technical solution, the gas concentration monitoring data set and energy consumption monitoring data set of the variable air volume fume hood are first obtained; then, the gas concentration monitoring data set and energy consumption monitoring data set of the variable air volume fume hood are respectively input into the preset harmful gas concentration dynamic prediction model and energy consumption prediction model, and the harmful gas concentration dynamic prediction value and energy consumption prediction value are output; finally, with the minimum energy consumption as the goal, considering the harmful gas concentration safety threshold constraint, the goal is solved based on the genetic algorithm to obtain the optimal face wind speed sequence of the variable air volume fume hood. This process solves the fume hood face wind speed control sequence with the lowest energy consumption under the premise of strictly ensuring that the harmful gas concentration is always below the safety threshold, thereby achieving the coordinated optimization of the operational safety and energy efficiency of the laboratory fume hood. While ensuring safety, the present invention significantly improves energy-saving performance.

[0012] Preferably, the gas concentration monitoring data set D c Including: surface wind speed sequence V, temperature sequence T, humidity sequence H, harmful gas concentration sequence C;

[0013] The preset harmful gas concentration dynamic prediction model F transformerIt includes the sequentially connected input embedding layer, position encoding layer, multi-head self-attention layer, feedforward neural network and output layer;

[0014] Among them, the input embedding layer is used to receive the surface wind speed v at time k k , temperature t k 、Humidity k and harmful gas concentration c k , opposite wind speed v k , temperature t k 、Humidity k and harmful gas concentration c k After linear transformation, they are mapped to high-dimensional space and finally concatenated into a unified feature vector E;

[0015] The position encoding layer is used to superimpose the time step information into the feature vector E to form the position encoding vector E ′ ;

[0016] The multi-head self-attention layer is used to encode the position vector E ′ Perform weighted interactions to capture long-term dependencies between different time steps and form a high-dimensional vector x;

[0017] The feedforward neural network is used to perform nonlinear transformation on the high-dimensional vector x output by the multi-head self-attention layer, further extract deep features, form an output vector, and output it to the output layer;

[0018] The output layer is used to perform linear regression on the output vector processed by the feedforward neural network to obtain the predicted value of the harmful gas concentration at time k+1.

[0019] Preferably, the parameter of the harmful gas concentration dynamic prediction model is θ1, and an objective function is constructed to optimize the parameter θ1 of the harmful gas concentration dynamic prediction model. The expression of the objective function is:

[0020]

[0021] Among them, c i Indicates the actual value of the harmful gas concentration, Represents the predicted value of harmful gas concentration, and N is the number of harmful gas concentration values involved in the prediction;

[0022] When the objective function reaches the minimum value, the objective function converges and the optimal parameters of the dynamic prediction model of harmful gas concentration are obtained.

[0023] Preferably, the energy consumption monitoring data set D e Includes: variable time surface wind speed sequence U, energy consumption E;

[0024] The energy consumption prediction model F LSTMIt includes sequentially connected sequence encoding layer, attention mechanism layer and feature fusion layer;

[0025] Among them, the sequence encoding layer is used to process the variable time surface wind speed sequence U using a bidirectional LSTM network to obtain a bidirectional feature vector h that integrates the forward hidden state and the reverse hidden state of the variable time surface wind speed sequence U at time step t. t ;

[0026] The attention mechanism layer is used to generate the bidirectional feature vector h based on the output of the sequence encoding layer. t Calculate its attention weight α t , forming the context vector c;

[0027] The feature fusion layer is used to implement the nonlinear transformation of the context vector c using the ReLU activation function, and then pass through the linear output layer to obtain the energy consumption prediction value.

[0028] Preferably, an objective function is constructed to optimize the parameter θ2 of the energy consumption prediction model, and the expression is:

[0029]

[0030] Where λ is the regularization coefficient, Used to calculate energy consumption forecast and actual energy consumption E i The error between them can be expressed as:

[0031]

[0032] Where N is the number of energy consumption values involved in the prediction;

[0033] When the objective function reaches the minimum value, the optimization process converges and the optimal parameters of the energy consumption prediction model are obtained.

[0034] Preferably, the face velocity sequence individuals of the variable air volume fume hood are used as the population of the genetic algorithm, and the process of calculating the fitness value of the population based on the predicted value of the harmful gas concentration and the predicted value of the energy consumption is as follows:

[0035] The surface wind speed sequence is used as the population of the genetic algorithm and initialized to generate S initial surface wind speed sequence individuals V (s) , the expression is:

[0036]

[0037] Where, s=1,2,…,S, represents the surface wind speed value of the sth surface wind speed sequence individual at time node i;

[0038] Get real-time sensor monitoring data D now, including the real-time temperature T now , Real-time harmful gas concentration C now , real-time humidity H now ;

[0039] Based on each surface wind speed sequence individual V (s) , using the harmful gas concentration dynamic prediction model to obtain the surface wind speed sequence individual V (s) The corresponding predicted value of harmful gas concentration at time r+1 The expression is:

[0040]

[0041] The energy efficiency prediction model is used to obtain the individual V of the surface wind speed sequence (s) The corresponding energy consumption forecast value at time r+1 The expression is:

[0042]

[0043] Set the safety threshold of harmful gas concentration to c eps , construct the fitness calculation function of the surface wind speed sequence individual, the expression is:

[0044]

[0045] Among them, β is the penalty coefficient.

[0046] Preferably, with the goal of minimizing energy consumption and considering the safety threshold constraint of harmful gas concentration, the process of solving the goal based on the genetic algorithm includes:

[0047] Select an operation process:

[0048] For the population P = {V (1) ,…,V (S)}S times of sampling with replacement, the sth surface wind speed sequence individual V in population P (s) Probability of being sampled The expression is:

[0049]

[0050] Among them, min(Fitness) is the minimum fitness value of the individual in the surface wind speed sequence in the current population;

[0051] After sampling the population, the mating pool P is obtained mating ={V (p1) ,…,V (ps)}, where V (pi) represents the parent individual selected in the i-th sampling;

[0052] Crossover operation process:

[0053] Randomly select two parent face wind speed sequence individuals in the mating pool to form a parent face wind speed sequence individual pair, cross-combine some face wind speed values in the parent face wind speed sequence individuals, and obtain the offspring face wind speed sequence individual V (c1) With V (c2) , the expression is:

[0054]

[0055] Among them, k cross are uniformly distributed random intersection points;

[0056] The offspring surface wind speed sequence individuals are integrated to obtain the offspring surface wind speed sequence individual set P child , the expression is:

[0057] P child ={V (c1) ,…,V (cS)}

[0058] Among them, V (ci) Indicates the i-th offspring individual obtained;

[0059] Mutation operation process:

[0060] Set the mutation probability parameter P m , traverse the individual set P of offspring surface wind speed sequence child The surface wind speed value of each individual in the surface wind speed sequence is expressed as P m The probability of mutation is expressed as:

[0061]

[0062] in, Represents the offspring surface wind speed sequence individual V (c) The surface wind speed value at the i-th time node, N(0,σ 2 ) is Gaussian perturbation;

[0063] The selection, crossover and mutation operations are used in turn to iteratively update the surface wind speed sequence population. When the number of cyclic iterations reaches the set total number of iterations or the difference in the minimum energy consumption of the surface wind speed sequence individuals of two adjacent generations is less than eps, the iteration is terminated and the optimal surface wind speed sequence and minimum energy consumption are obtained.

[0064] In a second aspect, the present application also proposes a variable air volume fume hood energy consumption control system based on face velocity optimization, the system comprising:

[0065] A monitoring data acquisition module is used to obtain the gas concentration monitoring data set and energy consumption monitoring data set of the variable air volume fume hood;

[0066] The prediction module is used to input the gas concentration monitoring data set of the variable air volume fume hood into the preset dynamic prediction model of harmful gas concentration and output the predicted value of harmful gas concentration; input the energy consumption monitoring data set into the preset energy consumption prediction model and output the predicted value of energy consumption;

[0067] The optimization module is used to take minimum energy consumption as the goal, consider the safety threshold constraint of harmful gas concentration, solve the goal based on genetic algorithm, and obtain the optimal face wind speed sequence of variable air volume fume hood; among them, the face wind speed sequence individuals of the variable air volume fume hood are used as the population of the genetic algorithm, and the fitness value of the population is calculated based on the predicted value of harmful gas concentration and the predicted value of energy consumption.

[0068] On the third aspect, the present application also proposes a variable air volume fume hood energy consumption control device based on surface wind speed optimization. The device includes a memory, a processor, and a computer program stored in the memory that can be run by the processor. The processor executes the computer program to implement the variable air volume fume hood energy consumption control method based on surface wind speed optimization.

[0069] In a fourth aspect, the present application also proposes a computer-readable storage medium on which a computer program is stored. The computer program includes program instructions. When the program instructions are executed by a computer, the computer executes the variable air volume fume hood energy consumption control method based on surface wind speed optimization.

[0070] Compared with the prior art, the present invention has the following beneficial effects:

[0071] This invention proposes a variable air volume fume hood energy consumption control method and system based on face velocity optimization. The method first obtains a gas concentration monitoring dataset and an energy consumption monitoring dataset for the variable air volume fume hood. The gas concentration monitoring dataset and energy consumption monitoring dataset are then input into a preset dynamic prediction model for harmful gas concentration and an energy consumption prediction model, respectively, to output dynamic prediction values for harmful gas concentration and energy consumption. Finally, with minimal energy consumption as the goal and taking into account the safety threshold constraint for harmful gas concentration, a genetic algorithm is used to solve the problem and obtain the optimal face velocity sequence for the variable air volume fume hood. This process solves the fume hood face velocity control sequence with the lowest energy consumption, while strictly ensuring that the harmful gas concentration remains below the safety threshold. This achieves the coordinated optimization of the operational safety and energy efficiency of the laboratory fume hood. While ensuring safety, the present invention significantly improves energy-saving performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 A schematic flow chart showing a method for controlling energy consumption of a variable air volume fume hood based on face velocity optimization proposed in Example 1 of the present invention;

[0073] Figure 2A schematic diagram showing the structure of the harmful gas concentration prediction model proposed in Example 2 of the present invention;

[0074] Figure 3 A schematic diagram showing the structure of the energy consumption prediction model proposed in Example 2 of the present invention;

[0075] Figure 4 A schematic diagram showing the structure of a variable air volume fume hood energy consumption control system based on face velocity optimization proposed in Example 4 of the present invention;

[0076] Figure 5 A schematic diagram showing the structure of the energy consumption control device for a variable air volume fume hood based on face velocity optimization proposed in Example 5 of the present invention; DETAILED DESCRIPTION

[0077] The accompanying drawings are for illustrative purposes only and are not to be construed as limiting this patent;

[0078] In order to better illustrate this embodiment, some parts of the drawings may be omitted, enlarged, or reduced, and do not represent the actual size;

[0079] It is understandable to those skilled in the art that descriptions of certain well-known contents may be omitted in the drawings.

[0080] The technical solution of the present invention is further described below with reference to the accompanying drawings and embodiments.

[0081] The positional relationships described in the drawings are for illustrative purposes only and should not be construed as limiting this patent;

[0082] Example 1

[0083] This embodiment proposes a variable air volume fume hood energy consumption control method based on face velocity optimization. The flow chart of this method is shown in FIG. Figure 1 , including the following steps:

[0084] S1. Obtain the gas concentration monitoring dataset and energy consumption monitoring dataset of the variable air volume fume hood;

[0085] S2 inputs the gas concentration monitoring data set of the variable air volume fume hood into a preset dynamic prediction model for harmful gas concentration and outputs a predicted value of the harmful gas concentration; inputs the energy consumption monitoring data set into a preset energy consumption prediction model and outputs a predicted value of energy consumption;

[0086] S3. With the goal of minimizing energy consumption and considering the safety threshold constraint of harmful gas concentration, the goal is solved based on the genetic algorithm to obtain the optimal face wind speed sequence of the variable air volume fume hood; among which, the face wind speed sequence individuals of the variable air volume fume hood are used as the population of the genetic algorithm, and the fitness value of the population is calculated based on the predicted value of the harmful gas concentration and the predicted value of energy consumption.

[0087] In this embodiment, the gas concentration monitoring dataset and energy consumption monitoring dataset of the variable air volume fume hood are first obtained; then, the gas concentration monitoring dataset and energy consumption monitoring dataset of the variable air volume fume hood are respectively input into a preset harmful gas concentration dynamic prediction model and energy consumption prediction model, and the harmful gas concentration dynamic prediction value and energy consumption prediction value are output; finally, with the minimum energy consumption as the goal, considering the harmful gas concentration safety threshold constraint, the goal is solved based on the genetic algorithm to obtain the optimal face wind speed sequence of the variable air volume fume hood. This process solves the fume hood face wind speed control sequence with the lowest energy consumption under the premise of strictly ensuring that the harmful gas concentration is always below the safety threshold, thereby achieving the coordinated optimization of the operational safety and energy efficiency of the laboratory fume hood. While ensuring safety, the present invention significantly improves energy-saving performance.

[0088] Example 2

[0089] In this embodiment, the gas concentration monitoring data set D c Including: surface wind speed sequence V, temperature sequence T, humidity sequence H, harmful gas concentration sequence C;

[0090] The preset harmful gas concentration dynamic prediction model F transformer It includes the input embedding layer, position encoding layer, multi-head self-attention layer, feedforward neural network and output layer connected in sequence; the structural diagram of the dynamic prediction model of harmful gas concentration is as follows Figure 2 As shown;

[0091] Among them, the input embedding layer is used to receive the surface wind speed v at time k k , temperature t k 、Humidity k and harmful gas concentration c k , opposite wind speed v k , temperature t k 、Humidity k and harmful gas concentration c k After linear transformation, they are mapped to high-dimensional space respectively and finally concatenated into a unified feature vector E;

[0092] The position encoding layer is used to superimpose the time step information into the feature vector E to form the position encoding vector E ′ ;

[0093] The multi-head self-attention layer is used to encode the position vector E ′ Perform weighted interactions to capture long-term dependencies between different time steps and form a high-dimensional vector x;

[0094] The feedforward neural network is used to perform nonlinear transformation on the high-dimensional vector x output by the multi-head self-attention layer, further extract deep features, form an output vector, and output it to the output layer;

[0095] The output layer is used to perform linear regression on the output vector processed by the feedforward neural network to obtain the predicted value of the harmful gas concentration at time k+1.

[0096] Specifically, the gas concentration monitoring data set in this embodiment contains 60,000 time series data, with a collection interval of 30 minutes. Each data contains a 30-minute time series of wind speed, temperature, humidity, and harmful gas concentration.

[0097] Specifically, the input embedding layer receives the surface wind speed v at time k k , temperature t k 、Humidity k and harmful gas concentration c k As input, these original features are linearly transformed according to the following formula and mapped into a high-dimensional space and concatenated into a unified feature vector E:

[0098] e v =W v v k +b v ,

[0099] e t =W t t k +b t ,

[0100] e h =W h h k +b h ,

[0101] e c =W c c k +b c ,

[0102] E=Concat(e v ,e t ,e h ,e c )

[0103] Among them, W v 、W t 、W h 、W c is the learnable weight matrix, b v 、b t 、b h 、b c is the learnable bias matrix, Concat(e v ,e t ,e h ,e c) means to convert e v ,e t ,e h ,e c Splicing into a vector; this step maps low-dimensional features to high-dimensional space through linear transformation, significantly enhancing feature expression capabilities. The splicing operation achieves multi-feature fusion, enabling the model to automatically learn the cross-modal coupling relationship of wind speed, temperature, humidity, and concentration;

[0104] Specifically, the position encoding layer superimposes the time step information on the feature vector E in the form of sine function and cosine function to form a position encoding vector E ′ , which enables the model to perceive the temporal relationship, and the expression is:

[0105]

[0106] E ′ =E+PE

[0107] Where pos represents the absolute position index of the current time step, i represents the feature dimension index, and PE represents the temporal position encoding generated based on the time step information. This step captures both the absolute position and the relative temporal relationship through the sine / cosine alternating encoding mechanism, enabling the model to capture both short-term local fluctuations and long-term trend changes.

[0108] Specifically, the multi-head self-attention layer encodes the position vector E ′ Perform weighted interactions to automatically capture long-term dependencies between different time steps. The expression is:

[0109] Q=E ′ W Q ,K=E ′ W K ,V=E ′ W V

[0110]

[0111] MultiHead(Q,K,V)=Concat(head1,…,head h )W o

[0112] Where Q is the query matrix, which represents the attention requirements of the current time step, K is the key matrix, which is used to match the reference features of the query, V is the value matrix, which represents the actual feature information, and W Q 、W K 、W V is the learnable query, key, and value projection matrix, d kRepresents the single-head attention dimension, softmax is an exponential function normalized by row, used to generate attention weights, head i represents the i-th attention head, W o Represents the weight of projecting the spliced multi-head output back to the original feature space. This step uses a parallel multi-head mechanism to learn the interaction patterns of different physical quantities such as wind speed-concentration and temperature-humidity, thereby enhancing the model's ability to analyze complex environmental coupling relationships.

[0113] Specifically, the feedforward neural network performs a nonlinear transformation on the output of the multi-head self-attention layer to further extract deep features, as shown in the following formula:

[0114] FFN(x)=ReLU(xW1+b1)W2+b2

[0115] Where W1 and W2 are the weight matrices of the feedforward layer, b1 and b2 represent the bias vectors of the feedforward layer, and ReLU is the nonlinear activation function. FFN() represents the final output of the feedforward neural network.

[0116] Specifically, the output layer uses linear regression to obtain the predicted value of the harmful gas concentration at time k+1. The expression is:

[0117]

[0118] Where W out and b out Represent the output weight matrix and output bias term respectively;

[0119] In this embodiment, the parameter of the harmful gas concentration dynamic prediction model is θ1, and an objective function is constructed to optimize the parameter θ1 of the harmful gas concentration dynamic prediction model. The expression of the objective function is:

[0120]

[0121] Among them, c i Indicates the actual value of the harmful gas concentration, Represents the predicted value of harmful gas concentration, and N is the number of harmful gas concentration values involved in the prediction;

[0122] When the objective function reaches the minimum value, the objective function converges and the optimal parameters of the dynamic prediction model of harmful gas concentration are obtained.

[0123] Specifically, the parameters of the dynamic prediction model for harmful gas concentration include the learnable weight matrix W of the input embedding layer v 、W t 、W h 、W c, the learnable bias matrix b v 、b t 、b h 、b c ; The query projection matrix W of the multi-head self-attention layer Q , key projection matrix W K , value projection matrix W V ; The weight W of the spliced multi-head output projected back to the original feature space o ; The feedforward layer weight matrices W1 and W2 of the feedforward neural network, the feedforward layer bias vectors b1 and b2 of the feedforward neural network; the output weight matrix W of the output layer out , the output bias term b of the output layer out .

[0124] In this embodiment, the energy consumption monitoring data set D e Includes: variable time surface wind speed sequence U, energy consumption E;

[0125] The energy consumption prediction model F LSTM It includes sequentially connected sequence encoding layer, attention mechanism layer and feature fusion layer;

[0126] Among them, the sequence encoding layer is used to process the variable time surface wind speed sequence U using a bidirectional LSTM network to obtain a bidirectional feature vector h that combines the forward hidden state and the reverse hidden state of the variable time surface wind speed sequence U. t ;

[0127] The attention mechanism layer is used to generate the bidirectional feature vector h based on the output of the sequence encoding layer. t Calculate its attention weight α t , forming the context vector c;

[0128] The feature fusion layer is used to implement the nonlinear transformation of the context vector c using the ReLU activation function, and then pass through the linear output layer to obtain the energy consumption prediction value.

[0129] Specifically, the energy consumption prediction model F LSTM The sequence modeling architecture uses a bidirectional long short-term memory network combined with an attention mechanism, including:

[0130] The sequence encoding layer is used to process the variable time surface wind speed sequence using a three-layer bidirectional LSTM network. Each layer contains 128 hidden units, and the Dropout mechanism is used to prevent overfitting. The hidden state calculation formula for each time step t is:

[0131]

[0132] in, represents the hidden state of the forward LSTM at time step t,

[0133] represents the hidden state of the reverse LSTM at time step t, h t Represents a bidirectional feature vector formed by concatenating the forward and reverse hidden states. This step uses the forward LSTM to capture historical wind speed variation patterns and the reverse LSTM to extract future wind speed trend features, enhancing the ability to analyze non-stationary time series.

[0134] The attention mechanism layer is used to calculate the attention weights for the hidden state sequence output by LSTM, adaptively focus on the key time step information, and p}Calculate attention weight α t And the formula for forming the context vector c is as follows:

[0135]

[0136] Among them, W h , b h and q are both learnable parameters;

[0137] The feature fusion layer is used to input the attention-weighted feature vector into the four-layer fully connected network, and realize nonlinear transformation through the ReLU activation function, and finally obtain the energy consumption prediction value through the linear output layer. The corresponding formula is as follows:

[0138] z1=ReLU(W1c+b1)

[0139] z2=ReLU(W2z1+b2)

[0140] z3=ReLU(W3z2+b3)

[0141]

[0142] Among them, W1, W2, W3, W out , b1, b2, b3 and b out All of them are learnable parameters. This step fits the complex mapping relationship between wind speed and energy consumption through nonlinear feature extraction, and retains high-order features that are strongly related to energy consumption. The structural diagram of the energy consumption prediction model is shown in the figure. Figure 3 shown.

[0143] In this embodiment, an objective function is constructed to optimize the parameter θ2 of the energy consumption prediction model, and the expression is:

[0144]

[0145] Where λ is the regularization coefficient, Used to calculate energy consumption forecast and actual energy consumption E iThe error between them can be expressed as:

[0146]

[0147] Where N is the number of energy consumption values involved in the prediction;

[0148] When the objective function reaches the minimum value, the optimization process converges and the optimal parameters of the energy consumption prediction model are obtained.

[0149] Specifically, the parameters θ2 of the energy consumption prediction model include: the learnable parameters W of the attention mechanism layer h , b h and q; the learnable weight matrices W1, W2, W3, W of the fully connected network of the feature fusion layer out , the learnable bias matrices b1, b2, b3 and b of the fully connected network of the feature fusion layer out .

[0150] This embodiment drives the model to learn real physical laws by minimizing the deviation between predicted energy consumption and actual energy consumption, and prevents overfitting and improves the model's generalization ability for new working condition wind speed sequences by adding an L2 regularization term to constrain the parameter norm.

[0151] In this embodiment, the face velocity sequence individuals of the variable air volume fume hood are used as the population of the genetic algorithm, and the process of calculating the fitness value of the population based on the predicted value of the harmful gas concentration and the predicted value of the energy consumption is as follows:

[0152] The surface wind speed sequence is used as the population of the genetic algorithm and initialized to generate S initial surface wind speed sequence individuals V (s) , the expression is:

[0153]

[0154] Where, s=1,2,…,S, represents the surface wind speed value of the sth surface wind speed sequence individual at time node i;

[0155] Get real-time sensor monitoring data D now , including the real-time temperature T now , Real-time harmful gas concentration C now , real-time humidity H now ;

[0156] Based on each surface wind speed sequence individual V (s) , using the harmful gas concentration dynamic prediction model to obtain the surface wind speed sequence individual V (s) The corresponding predicted value of harmful gas concentration at time r+1 The expression is:

[0157]

[0158] The energy efficiency prediction model is used to obtain the individual V of the surface wind speed sequence (s) The corresponding energy consumption forecast value at time r+1 The expression is:

[0159]

[0160] Set the safety threshold of harmful gas concentration to c eps , construct the fitness calculation function of the surface wind speed sequence individual, the expression is:

[0161]

[0162] Among them, β is the penalty coefficient.

[0163] This embodiment converts the energy consumption minimization problem into a fitness maximization problem, and softly constrains the safe concentration of harmful gases through a quadratic penalty term, so that when the predicted concentration of harmful gases exceeds the safe threshold of harmful gas concentration, an exponential penalty is imposed, thereby determining which individuals can enter the mating pool and controlling the subsequent evolutionary direction of the population.

[0164] In this embodiment, with the goal of minimizing energy consumption and considering the safety threshold constraint of harmful gas concentration, the process of solving the goal based on the genetic algorithm includes:

[0165] Select an operation process:

[0166] For the population P = {V (1) ,…,V (S)}S times of sampling with replacement, the sth surface wind speed sequence individual V in population P (s) Probability of being sampled The expression is:

[0167]

[0168] Among them, min(Fitness) is the minimum fitness value of the individual in the surface wind speed sequence in the current population;

[0169] After sampling the population, the mating pool P is obtained mating ={V (p1) ,…,V (pS)}, where V (pi) represents the parent individual selected in the i-th sampling;

[0170] Crossover operation process:

[0171] Randomly select two parent face wind speed sequence individuals in the mating pool to form a parent face wind speed sequence individual pair, cross-combine some face wind speed values in the parent face wind speed sequence individuals, and obtain the offspring face wind speed sequence individual V(c1) With V (c2) , the expression is:

[0172]

[0173] Among them, k cross are uniformly distributed random intersection points;

[0174] The offspring surface wind speed sequence individuals are integrated to obtain the offspring surface wind speed sequence individual set P child , the expression is:

[0175] P child ={V (c1) ,…,V (cS)}

[0176] Among them, V (ci) Indicates the i-th offspring individual obtained;

[0177] Mutation operation process:

[0178] Set the mutation probability parameter P m , traverse the individual set P of offspring surface wind speed sequence child The surface wind speed value of each individual in the surface wind speed sequence is expressed as P m The probability of mutation is expressed as:

[0179]

[0180] in, Represents the offspring surface wind speed sequence individual V (c) The surface wind speed value at the i-th time node, N(0,σ 2 ) is Gaussian perturbation;

[0181] The surface wind speed sequence population is updated by cyclic iteration using selection, crossover and mutation operations. When the number of cyclic iterations reaches the set total number of iterations or the difference in the minimum energy consumption of the surface wind speed sequence individuals of two adjacent generations is less than eps=10 -5 When , the iteration is terminated and the optimal surface wind speed sequence and minimum energy consumption are obtained.

[0182] Example 3

[0183] In this example, a comparative analysis was conducted to compare energy consumption between the optimal face wind speed sequence obtained through multi-objective genetic algorithm optimization and a traditional fixed face wind speed control strategy. 0.3 m / s, 0.4 m / s, 0.5 m / s, and 0.6 m / s were selected as fixed face wind speed baselines for the comparative analysis. Real-time sensor data was collected to form a set of 300 test samples. Each test sample contained continuous data on face wind speed, temperature, humidity, and hazardous gas concentrations for the current time.

[0184] The steps of the comparative experiment are:

[0185] (1) For each test sample, five different face wind speed control strategies were set: fixed face wind speeds of 0.3 m / s, 0.4 m / s, 0.5 m / s, and 0.6 m / s, and the optimal face wind speed sequence calculated by a multi-objective genetic algorithm;

[0186] (2) Input the surface wind speed sequence under each control strategy into the dynamic prediction model of harmful gas concentration to obtain the predicted value of harmful gas concentration at the next moment;

[0187] (3) Input the corresponding surface wind speed sequence into the energy consumption prediction model to calculate the corresponding predicted energy consumption value;

[0188] (4) Count the number of times and the compliance rate of the predicted value of harmful gas concentration below the safety threshold under each control strategy;

[0189] (5) Calculate the average energy consumption and relative energy saving rate of each control strategy. The comparison indicators include: safety compliance rate = (number of samples with concentration prediction value ≤ safety threshold / total number of samples) × 100%; relative energy saving rate = (comparison strategy energy consumption - baseline strategy energy consumption) / baseline strategy energy consumption × 100%; comprehensive performance index = safety compliance rate / normalized average energy consumption.

[0190] Specifically, the performance comparison of different control strategies for variable air volume fume hoods is shown in Table 1:

[0191] Table 1

[0192]

[0193] The method proposed in this application achieved a safety compliance rate of 97.0%, which is on par with the fixed wind speed 0.6m / s strategy, with a relative energy saving rate of -10.0%, and a comprehensive performance index of 88.18. In terms of energy-saving performance, the control method proposed in this application has significant advantages over the existing fixed wind speed strategy, far better than -25.0% of a fixed wind speed of 0.5m / s and -39.0% of a fixed wind speed of 0.6m / s. Under the premise of ensuring the same safety compliance rate, the energy-saving effect of the present invention is 29 percentage points higher than that of the fixed wind speed 0.6m / s strategy. The control strategy of this application performed most outstandingly in the comprehensive performance index, reaching 88.18, which is 16.18, 12.73, 15.38 and 18.40 percentage points higher than the fixed wind speed 0.3m / s, 0.4m / s, 0.5m / s and 0.6m / s strategies respectively. In summary, the variable air volume control strategy of the present invention achieves an optimal balance between safety and energy saving through dynamic adjustment, and achieves better results in various performance indicators.

[0194] Example 3

[0195] This embodiment proposes a variable air volume fume hood energy consumption control system based on face velocity optimization. In this embodiment, the system is used to implement a variable air volume fume hood energy consumption control method based on face velocity optimization. The structural diagram is shown in FIG. Figure 4 Shown, including:

[0196] A monitoring data acquisition module is used to obtain the gas concentration monitoring data set and energy consumption monitoring data set of the variable air volume fume hood;

[0197] The prediction module is used to input the gas concentration monitoring data set of the variable air volume fume hood into the preset dynamic prediction model of harmful gas concentration and output the predicted value of harmful gas concentration; input the energy consumption monitoring data set into the preset energy consumption prediction model and output the predicted value of energy consumption;

[0198] The optimization module is used to take minimum energy consumption as the goal, consider the safety threshold constraint of harmful gas concentration, solve the goal based on genetic algorithm, and obtain the optimal face wind speed sequence of variable air volume fume hood; among them, the face wind speed sequence individuals of the variable air volume fume hood are used as the population of the genetic algorithm, and the fitness value of the population is calculated based on the predicted value of harmful gas concentration and the predicted value of energy consumption.

[0199] Example 4

[0200] This embodiment proposes a variable air volume fume hood energy consumption control device based on face velocity optimization. The device includes a memory 101, a processor 102, and a computer program stored in the memory 101 and executable by the processor. The processor 102 executes the computer program to implement a variable air volume fume hood energy consumption control method based on face velocity optimization. The structural diagram of the variable air volume fume hood energy consumption control device based on face velocity optimization is shown in FIG. Figure 5 shown.

[0201] In another embodiment, a computer-readable storage medium is provided, on which a computer program is stored, the computer program including program instructions.

[0202] When the program instructions are executed by a computer, the computer is caused to execute a variable air volume fume hood energy consumption control method based on face velocity optimization.

[0203] Obviously, the above embodiments of the present invention are merely examples for the purpose of clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. Those skilled in the art will appreciate that other variations or modifications can be made based on the above description. It is not necessary and impossible to enumerate all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the claims of the present invention.

Claims

1. A variable air volume fume hood energy consumption control method based on face velocity optimization, characterized in that: The following steps are involved: Obtain gas concentration monitoring data sets and energy consumption monitoring data sets for variable air volume fume hoods; Input the gas concentration monitoring data set of the variable air volume fume hood into the preset dynamic prediction model of harmful gas concentration, and output the predicted value of harmful gas concentration; input the energy consumption monitoring data set into the preset energy consumption prediction model, and output the energy consumption prediction value; With the goal of minimizing energy consumption and considering the safety threshold constraint of harmful gas concentration, the genetic algorithm is used to solve the goal and obtain the optimal face wind speed sequence of variable air volume fume hoods. Among them, the face wind speed sequence individuals of the variable air volume fume hoods are used as the population of the genetic algorithm, and the fitness value of the population is calculated based on the predicted value of harmful gas concentration and the predicted value of energy consumption.

2. The energy consumption control method of a variable air volume fume hood based on face velocity optimization according to claim 1 is characterized in that: The gas concentration monitoring data set D c Including: surface wind speed sequence V, temperature sequence T, humidity sequence H and harmful gas concentration sequence C; The preset harmful gas concentration dynamic prediction model F transformer It includes the sequentially connected input embedding layer, position encoding layer, multi-head self-attention layer, feedforward neural network and output layer; Among them, the input embedding layer is used to receive the surface wind speed v at time k k , temperature t k 、Humidity k and harmful gas concentration c k , opposite wind speed v k , temperature t k 、Humidity k and harmful gas concentration c k After linear transformation, they are mapped to high-dimensional space and finally concatenated into a unified feature vector E; The position encoding layer is used to superimpose the time step information into the feature vector E to form the position encoding vector E′; The multi-head self-attention layer is used to perform weighted interactions on the position encoding vector E′ to capture the long-term dependencies between different time steps and form a high-dimensional vector x; The feedforward neural network is used to perform nonlinear transformation on the high-dimensional vector x output by the multi-head self-attention layer, further extract deep features, form an output vector, and output it to the output layer; The output layer is used to perform linear regression on the output vector processed by the feedforward neural network to obtain the predicted value of the harmful gas concentration at time k+1.

3. The energy consumption control method of a variable air volume fume hood based on face velocity optimization according to claim 2 is characterized in that: Assume that the parameter of the dynamic prediction model of harmful gas concentration is θ1, and construct an objective function to optimize the parameter θ1 of the dynamic prediction model of harmful gas concentration. The expression of the objective function is: Among them, c i Indicates the actual value of the harmful gas concentration, Represents the predicted value of harmful gas concentration, and N is the number of harmful gas concentration values involved in the prediction; When the objective function reaches the minimum value, the objective function converges and the optimal parameter θ1 of the dynamic prediction model of harmful gas concentration is obtained.

4. The energy consumption control method of a variable air volume fume hood based on face velocity optimization according to claim 1, characterized in that: The energy consumption monitoring dataset D e Including: variable time surface wind speed series U, energy consumption E; The energy consumption prediction model F LSTM It includes sequentially connected sequence encoding layer, attention mechanism layer and feature fusion layer; Among them, the sequence encoding layer is used to process the variable time surface wind speed sequence U using a bidirectional LSTM network to obtain a bidirectional feature vector h that integrates the forward hidden state and the reverse hidden state of the variable time surface wind speed sequence U at time step t. t ; The attention mechanism layer is used to generate the bidirectional feature vector h based on the output of the sequence encoding layer. t Calculate its attention weight α t , forming the context vector c; The feature fusion layer is used to implement the nonlinear transformation of the context vector c using the ReLU activation function, and then pass through the linear output layer to obtain the energy consumption prediction value.

5. The variable air volume fume hood energy consumption control method based on face velocity optimization according to claim 4 is characterized in that: Construct an objective function to optimize the parameter θ2 of the energy consumption prediction model, and the expression is: Where λ is the regularization coefficient, Used to calculate energy consumption forecast and actual energy consumption E i The error between them can be expressed as: Where N is the number of energy consumption values involved in the prediction; When the objective function reaches the minimum value, the objective function converges and the optimal parameter θ2 of the energy consumption prediction model is obtained.

6. A variable air volume fume hood energy consumption control method based on face velocity optimization according to any one of claims 1 to 5, characterized in that: The process of calculating the fitness value of the population based on the predicted value of harmful gas concentration and energy consumption is as follows: The surface wind speed sequence is used as the population of the genetic algorithm and initialized to generate S initial surface wind speed sequence individuals V( s ), the expression is: Where, s=1,2,…,S, represents the surface wind speed value of the sth surface wind speed sequence individual at time node i; Get real-time sensor monitoring data D now , including the real-time temperature T now , Real-time harmful gas concentration C now , real-time humidity H now ; Based on each surface wind speed sequence individual V( s ), using the dynamic prediction model of harmful gas concentration to obtain the individual V of the surface wind speed sequence (s) The corresponding predicted value of harmful gas concentration at time r+1 The expression is: The energy efficiency prediction model is used to obtain the individual V( s ) corresponding to the energy consumption forecast value at time r+1 The expression is: Set the safety threshold of harmful gas concentration to c eps , construct the fitness calculation function of the surface wind speed sequence individual, the expression is: Among them, β is the penalty coefficient.

7. The variable air volume fume hood energy consumption control method based on face velocity optimization according to claim 6 is characterized in that: With the goal of minimizing energy consumption and considering the safety threshold constraint of harmful gas concentration, the process of solving the goal based on the genetic algorithm includes: Select an operation process: For the population P = {V (1) ,…,V (S) }S times of sampling with replacement, the sth surface wind speed sequence individual V in population P (s) Probability of being sampled The expression is: Among them, min(Fitness) is the minimum fitness value of the individual in the surface wind speed sequence in the current population; After sampling the population, the mating pool P is obtained mating ={V (p1) ,…,V (pS )}, where V (pi) represents the parent individual selected in the i-th sampling; Crossover operation process: Randomly select two parent face wind speed sequence individuals in the mating pool to form a parent face wind speed sequence individual pair, cross-combine some face wind speed values in the parent face wind speed sequence individuals, and obtain the offspring face wind speed sequence individual V( c1 ) and V( c2 ), the expression is: Among them, k cross are uniformly distributed random intersection points; The offspring surface wind speed sequence individuals are integrated to obtain the offspring surface wind speed sequence individual set P child , the expression is: P child ={V (c1) ,…,V (cS) } Among them, V( ci ) represents the i-th offspring individual obtained; Mutation operation process: Set the mutation probability parameter P m , traverse the individual set P of offspring surface wind speed sequence child The surface wind speed value of each individual in the surface wind speed sequence is expressed as P m The probability of mutation is expressed as: in, Represents the offspring surface wind speed sequence individual V (c) The surface wind speed value at the i-th time node, N(0,σ 2 ) is Gaussian perturbation; The selection, crossover and mutation operations are used in turn to iteratively update the surface wind speed sequence population. When the number of cyclic iterations reaches the set total number of iterations or the difference in the minimum energy consumption of the surface wind speed sequence individuals of two adjacent generations is less than eps, the iteration is terminated and the optimal surface wind speed sequence and minimum energy consumption are obtained.

8. A variable air volume fume hood energy consumption control system based on face velocity optimization, characterized in that: The system is used to implement the variable air volume fume hood energy consumption control method based on face velocity optimization according to any one of claims 1 to 7, comprising: A monitoring data acquisition module is used to obtain the gas concentration monitoring data set and energy consumption monitoring data set of the variable air volume fume hood; The prediction module is used to input the gas concentration monitoring data set of the variable air volume fume hood into the preset dynamic prediction model of harmful gas concentration and output the predicted value of harmful gas concentration; input the energy consumption monitoring data set into the preset energy consumption prediction model and output the predicted value of energy consumption; The optimization module is used to take minimum energy consumption as the goal, consider the safety threshold constraint of harmful gas concentration, solve the goal based on genetic algorithm, and obtain the optimal face wind speed sequence of variable air volume fume hood; among them, the face wind speed sequence individuals of the variable air volume fume hood are used as the population of the genetic algorithm, and the fitness value of the population is calculated based on the predicted value of harmful gas concentration and the predicted value of energy consumption.

9. A variable air volume fume hood energy consumption control device based on face velocity optimization, characterized in that: The device includes a memory, a processor, and a computer program stored in the memory and executable by the processor. The processor executes the computer program to implement the variable air volume fume hood energy consumption control method based on surface velocity optimization as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that A computer program is stored thereon, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer is caused to execute the variable air volume fume hood energy consumption control method based on surface wind speed optimization according to any one of claims 1 to 7.

Images