Central air conditioning aggregate load prediction method and system

Abstract

The application discloses a central air conditioner aggregated load prediction method and system, and relates to the technical field of central air conditioner control, and comprises the following steps: receiving central air conditioner historical operation data, using Latin hypercube sampling (LHS) to randomly and uniformly extract samples from the central air conditioner historical operation data to obtain central air conditioner sample data, wherein the central air conditioner historical operation data comprises temperature, humidity and power consumption; inputting the central air conditioner sample data into a pre-established THBA-Bi-LSTM model to output central air conditioner aggregated load prediction results, wherein the pre-established THBA-Bi-LSTM model is obtained by optimizing the parameters of Bi-LSTM by using a THBA algorithm; the adjustable load of the central air conditioner system can be more efficiently managed and controlled, and the goals of energy saving, stable operation and optimized resource utilization are achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of central air conditioning control, in particular to a central air conditioning aggregated load prediction method and system. BACKGROUND

[0002] With the rapid development of social economy, problems such as energy shortage and environmental pollution are becoming increasingly serious. Under the promotion of the "double carbon" goal, the country is vigorously building a new energy system, and the proportion of new energy power generation is continuously increasing. However, new energy power generation has the characteristics of instability and intermittency, and a large number of access will bring frequency instability, serious harmonic and other power quality problems to the power grid. At the same time, the diversification of load makes the load side have great optimization potential, which will further improve the capacity of the power grid after being introduced into the power grid. In addition, the power load of the power grid is increasing year by year, and the proportion of temperature control load such as central air conditioning in the power load is high, and the contradiction between the power supply side and the demand side is becoming more and more prominent. Therefore, it is particularly important to promote high-precision prediction of central air conditioning aggregated load and optimize energy management to maintain the balance and stability of the power system.

[0003] In the field of central air conditioning control, there are some methods and systems for managing air conditioning load aggregation. For example, a method and device for evaluating the dynamic aggregation regulation potential of air conditioning load clusters, which is based on the equivalent thermal parameter model of air conditioning load, determines the cycle of air conditioning load according to outdoor temperature, upper limit of indoor temperature set value and lower limit of indoor temperature set value, and combines the relationship between air conditioning power and room temperature change. In addition, a control strategy for central air conditioner (CAC) load group participating in power system peak regulation, which proposes a central air conditioner single load model for participating in power system peak regulation, and further constructs a system architecture for central air conditioner load group participating in power system peak regulation, and proposes an optimization model for central air conditioner load group participating in power system peak regulation. The non-linear constraint is segmented and linearized, so that the model is convenient, fast and effective for solving. Most of the existing technologies focus on single regulation parameter load control, and lack accurate prediction methods for central air conditioning aggregated load. SUMMARY

[0004] To solve the problems mentioned in the background, the purpose of the present application is to provide a central air conditioning aggregated load prediction method and system, which can more efficiently manage and control the adjustable load of the central air conditioning system.

[0005] In the first aspect, the purpose of the present application can be realized by the following technical scheme: a central air conditioning aggregated load prediction method, the method comprising the following steps:

[0006] Receive central air conditioning historical operation data, and randomly and uniformly extract samples from the central air conditioning historical operation data by using a Latin hypercube sampling (LHS) method to obtain central air conditioning sample data, wherein the central air conditioning historical operation data includes temperature, humidity, and power consumption;

[0007] Input the central air conditioning sample data into a pre-established THBA-Bi-LSTM model, and output a central air conditioning aggregated load prediction result, wherein the pre-established THBA-Bi-LSTM model is obtained by optimizing parameters of a Bi-LSTM by using a THBA algorithm.

[0008] With reference to the first aspect, in some implementations of the first aspect, the method further includes: the Latin hypercube sampling (LHS) method simulates variation rules of air conditioner loads, and aggregates and superimposes the air conditioner loads.

[0009] With reference to the first aspect, in some implementations of the first aspect, the method further includes: the process of randomly and uniformly extracting samples from the central air conditioning historical operation data by using the Latin hypercube sampling (LHS) method includes the following steps:

[0010] Suppose that random variables X1, X2, …, X N have a cumulative probability distribution function Y N =F(X N ), first determine a sampling size N, and divide a probability distribution function into N parts uniformly;

[0011] Randomly sample once in each interval, and Y i =(r i -i-1) / N in the ith interval, where r i is a random number in [0, 1];

[0012] Substitute Y i into the inverse function F -1 of Y N =F(X N ), and obtain a sampling value X i of the interval i.

[0013] Repeat the above steps N times to generate N variables that satisfy the F distribution function.

[0014] With reference to the first aspect, in some implementations of the first aspect, the method further includes: the THBA algorithm is obtained by introducing a Tent chaotic mapping to improve initialization of a meerkat algorithm, and the THBA algorithm includes three stages, namely, an initialization stage, a mining stage, and a honey collecting stage.

[0015] With reference to the first aspect, in some implementations of the first aspect, the method further includes: a process of the initialization stage:

[0016] Set initialization parameters, including population size N, maximum iteration number T max , generate the initial meerkat population with chaotic characteristics using Tent mapping of formula (1);

[0017]

[0018] In the formula, X i is the mapping function value at the i-th iteration; γ is the chaotic coefficient; X i+1 is the mapping function value at the i+1-th iteration;

[0019] Calculate the meerkat fitness value, get the optimal meerkat fitness value, and determine the food source;

[0020] The process of the excavation phase: in the excavation phase, the meerkat performs a heart linear action, and the heart-shaped movement is simulated by formula (4):

[0021] Define the intensity I and the update density factor, I is the smell intensity of the prey, as shown in formula (2):

[0022]

[0023] Where s is the source intensity or the concentration intensity; d i represents the distance between the prey and the current meerkat individual, and r2 is a random number in (0, 1);

[0024] The density factor is updated using the method in formula (3):

[0025]

[0026] Where T max is the maximum iteration number; C≥1;

[0027] Formula (4) simulates the action of meerkat in the excavation phase:

[0028] x new =x p +F×β×I×x p +F×r3×α×d i ×|cos(2πr4)×[1-cos(2πr5)]| (4)

[0029] Where x p is the global optimal position in the current state; β≥1 is the ability of meerkat to obtain food, d i represents the distance between the prey and the current meerkat individual, r3, r4, r5 are three random numbers in (0, 1), and F is a flag to change the search direction, which is determined by formula (5):

[0030]

[0031] The honey harvesting phase; the case of honey badger following the honeyguide to the beehive is simulated by formula (6):

[0032] x new =x p +F×r7×α×d i (6)

[0033] Wherein, x new is the updated honey badger individual position; x p is the prey position; F and a are determined by formula (5) and formula (3) respectively; r7 is a random number in (0, 1), from formula (6), according to the distance information d i , the honey badger searches around the prey position x p .

[0034] In combination with the first aspect, in some implementations of the first aspect, the method further comprises: the process of optimizing the parameters of the Bi-LSTM by using the THBA algorithm comprises:

[0035] The central air conditioner aggregate power data is taken as an input characteristic variable, the initial parameters to be optimized are set and the initial values are given; the parameters to be optimized are the learning rate, the number of neurons in the first hidden layer, the number of neurons in the second hidden layer and the regularization coefficient;

[0036] In the training process, the root mean square error MSE of the predicted data and the actual data is used as the fitness function; the specific root mean square error MSE calculation formula is as follows:

[0037]

[0038] Wherein, y t is the actual aggregate load value at t time, is the predicted aggregate load value at t time, and N is the sample quantity;

[0039] The improved THBA algorithm is run to search and optimize the Bi-LSTM, and the parameters are adjusted to obtain the optimal solution.

[0040] In combination with the first aspect, in some implementations of the first aspect, the method further comprises: the Bi-LSTM comprises a forgetting gate, an input gate and an output gate, wherein the calculation formula of the forgetting gate is as follows:

[0041] f t =σ(W f [h t-1 ,x t ]+b f (8)

[0042] In the formula, W f is the weight matrix of the forget gate; b f is the bias variable of the forget gate, and σ is the activation function sigmoid.

[0043] The input gate is followed by the forget gate, and there is also a weight matrix W i in the input gate, and the bias value b i of the input gate is as follows:

[0044] i t = σ(W i [h t-1 ,x t ]+b i (9)

[0045] The current input of the new cell state r' t The calculation formula is as follows:

[0046] r' t = tanh(W r [h t-1 ,x t ]+b r (10)

[0047] The cell state r t at the current time, the calculation method is to multiply the forget gate f t at the current time and the cell state r t-1 at the last time, and then add the result of multiplying the current input i t at the current time and the current cell state r' t , and the calculation formula is as follows:

[0048] r t = f t ☉r t-1 +i t ☉r t (11)

[0049] The calculation of the output gate of the LSTM is as follows:

[0050] o t = σ(W o [h t-1 ,x t ]+b o ) (12)The calculation expression of the output gate of the LSTM, and the final output of the LSTM is related to the output gate and the cell state, and the calculation formula is as follows:

[0051] h t = o t ☉ tanh(r t ) (13)

[0052] With reference to the first aspect, in some implementations of the first aspect, the method further includes: the Bi-LSTM is to calculate an output value by a forward LSTM and an output value by a backward LSTM when the signal is transmitted from an input layer to a network layer, and the calculation process is as follows:

[0053]

[0054] wherein, x t is an input at time t, LSTM is a neural unit, is an output of a backward hidden layer at each time, is an output of a forward hidden layer at each time;

[0055] If the MSE reaches the minimum, a Bi-LSTM model is directly constructed by using the parameters to predict the aggregated power of the central air conditioner, otherwise, the next step is performed.

[0056] If the MSE does not reach the minimum, it is determined whether the maximum number of iterations is reached, if the maximum number of iterations is reached, a THBA-Bi-LSTM model is constructed by using the obtained parameters to realize the prediction of the aggregated load of the central air conditioner.

[0057] With reference to the first aspect, in some implementations of the first aspect, the method further includes: the process of inputting the central air conditioner sample data into the pre-established THBA-Bi-LSTM model and outputting the prediction result of the aggregated load of the central air conditioner includes:

[0058] Data preprocessing: the input data is normalized to the interval [0, 1] to improve the training efficiency of the model, and the normalization formula is as follows:

[0059]

[0060] wherein, x is the original data, x' is the normalized data, x min is the minimum value in the original data, x max is the maximum value in the original data, and the original data is the feature data related to the aggregated load.

[0061] The Bi-LSTM model structure, specifically, the input layer: receives input data such as temperature, humidity, and power consumption.

[0062] The Bi-LSTM layer: includes a forward LSTM layer and a backward LSTM layer, which respectively process the forward and backward information of the sequence data. The forward LSTM layer calculates the forget gate, the input gate, the new cell state, the cell state update, and the output gate, and the backward LSTM layer is similar to the forward LSTM layer, but processes the reverse sequence of the input data.

[0063] The aggregated output layer outputs the predicted aggregated load value, that is, the outputs of the forward and reverse LSTM layers are combined, and the final prediction result is output through a fully connected layer, and the aggregated load prediction follows the formula:

[0064]

[0065] In the formula, W y is the output layer weight matrix, and b y is the output layer bias vector, is the hidden state of the forward LSTM, is the hidden state of the reverse LSTM, is the aggregated load prediction output value at t.

[0066] The calculation process of the forward LSTM layer and the reverse LSTM layer in the Bi-LSTM model is described in formulas (7)-(15).

[0067] In the second aspect, to achieve the above object, the application discloses a central air conditioner aggregated load prediction system, comprising:

[0068] A data processing module is configured to receive central air conditioner historical operation data, randomly and uniformly extract samples from the central air conditioner historical operation data by using a Latin hypercube sampling (LHS) method, and obtain central air conditioner sample data, wherein the central air conditioner historical operation data includes temperature, humidity, and power consumption.

[0069] A load prediction module is configured to input the central air conditioner sample data into a pre-established THBA-Bi-LSTM model, and output a central air conditioner aggregated load prediction result, wherein the pre-established THBA-Bi-LSTM model is obtained by optimizing the parameters of the Bi-LSTM using a THBA algorithm.

[0070] The application has the following advantages:

[0071] Firstly, the Latin hypercube sampling (LHS) is used for central air conditioner aggregated modeling, which can realize fine management of the adjustable load of the central air conditioner, improve the control accuracy and stability of the system, and improve the comfort of the indoor environment by improving the coverage and representativeness of the samples, making the model more accurate and reliable, reducing system fluctuations, and improving the comfort of the indoor environment.

[0072] Secondly, the adaptability of the central air conditioner system in commercial scenarios is enhanced. Whether it is daily operation, peak electricity consumption or emergency demand response, the adjustable load can be quickly adjusted to meet the requirements of the system and improve the reliability and flexibility of the system.

[0073] Finally, the effect of the prediction model is improved by optimizing Bi-LSTM with the improved meleset algorithm THBA, so as to realize effective prediction of the central air conditioning load. Significantly reduce energy consumption, achieve the goal of energy saving and emission reduction. Accurate prediction of the load of central air conditioning, reasonable adjustment of load distribution, reduce unnecessary energy waste, contribute to sustainable development. BRIEF DESCRIPTION OF DRAWINGS

[0074] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced as follows. Obviously, for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings;

[0075] Figure 1 is a method flowchart of the present application;

[0076] Figure 2 is a central air conditioning aggregated load prediction process flowchart based on the THBA-Bi-LSTM model of the present application;

[0077] Figure 3 is a system structure diagram of the present application. DETAILED DESCRIPTION

[0078] The technical solutions in the embodiments of the present application will be described clearly and completely below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0079] Embodiment one:

[0080] Next, the related terms involved in the embodiments of the present application are introduced:

[0081] Load prediction: load prediction is to determine the load data at a certain time in the future according to the operating characteristics of the system, capacity expansion decision, natural conditions and social influence and many other factors, where load refers to power demand (power) or electricity consumption; Load prediction is an important content in economic dispatching of power system, and is an important module of energy management system (EMS).

[0082] As shown in Figure 1 , a central air conditioning aggregated load prediction method, the method comprising the following steps:

[0083] S101: Receive central air conditioning historical operation data, use Latin hypercube sampling (LHS) method to randomly and uniformly extract samples from the central air conditioning historical operation data, and obtain central air conditioning sample data, wherein the central air conditioning historical operation data includes temperature, humidity and power consumption;

[0084] Among them, the historical operation data of the central air conditioner in the commercial scene is collected and input, including but not limited to temperature, humidity, power consumption and the like; the variables and their distributions of the sampling are determined according to the data characteristics.

[0085] In modern commercial buildings, in order to improve the operation efficiency and management level of the central air conditioning system, efficient data acquisition technology becomes particularly important. Although the traditional data acquisition method such as manual recording has low cost, it has the problems of poor accuracy and large workload; while the sensor network and SCADA system improve the accuracy and real-time of data, but the initial investment is high and the maintenance is complex. Therefore, the present application uses the Internet of Things platform as the main means of central air conditioning system operation data acquisition, aiming to realize accurate monitoring and management of system operation state through automatic mode.

[0086] The data acquisition method of the present application firstly involves installing temperature and humidity sensors, electric energy meters and other devices at each key part of the central air conditioning system to ensure that they can accurately reflect the operation state of the system. Then, use wired or wireless communication technology to connect these devices to the Internet of Things platform to ensure the stability and reliability of data transmission. Then, set appropriate data acquisition frequency, such as once per minute or once per hour, upload the collected data to the cloud server or local data center for storage and management.

[0087] Further, the present application uses big data analysis and artificial intelligence technology to clean, process and analyze the collected data, generates visual reports to help managers make more scientific decisions. This method not only realizes high-precision and real-time monitoring of data, but also reduces the need for manual intervention through an automated data acquisition and management system, thereby reducing operating costs. In addition, remote monitoring and control support allows managers to view system operation status at any time and promptly discover and solve problems.

[0088] Finally, the Internet of Things platform data acquisition method provided by the present application has good scalability and flexibility, and can easily add new sensors and devices according to demand, and at the same time, the system is easy to maintain, and most problems can be solved through remote diagnosis and software updates. Modern Internet of Things platforms are usually equipped with advanced security mechanisms to ensure the security of data transmission and storage. In summary, the data acquisition method based on the Internet of Things platform is the best choice for central air conditioning system operation data acquisition in commercial scenarios, which can effectively improve the operation efficiency and management level of the system.

[0089] The Latin hypercube sampling LHS method simulates the variation law of each air conditioning load power, and aggregates and superimposes the air conditioning load.

[0090] The process of randomly and uniformly sampling the central air conditioning historical operation data by using the Latin hypercube sampling LHS method comprises the following steps:

[0091] Suppose that random variables X1, X2, …, X N The cumulative probability distribution function is Y N =F(X N ). First, determine the sampling size N, and divide the probability distribution function into N parts uniformly;

[0092] Random sampling is performed in each interval, such as Y i =(r i -i-1) / N in the ith interval, where r i is a random number in [0, 1];

[0093] Substitute Y i into the inverse function F N of Y N =F(X -1 ), and the sampling value X i of the interval i is obtained.

[0094] Repeat the above steps N times to generate N variables subject to the F distribution function.

[0095] The aggregated power characteristics of 3000 commercial air conditioners are obtained, and the data of the aggregated power of the central air conditioner in the commercial scenario are obtained.

[0096] S102: input the central air conditioner sample data into the pre-established THBA-Bi-LSTM model, and output the central air conditioner aggregated load prediction result, wherein the pre-established THBA-Bi-LSTM model is obtained by optimizing the parameters of Bi-LSTM using THBA algorithm.

[0097] The THBA algorithm is obtained by introducing Tent chaotic mapping to improve the initialization population of the meerkat algorithm, and the THBA algorithm includes three stages: initialization stage, mining stage and honey collecting stage.

[0098] Initialization stage; set initialization parameters such as population size N, maximum iteration number T max , etc., and generate the initial meerkat population with chaotic characteristics by using formula (1) Tent mapping;

[0099]

[0100] In the formula, X iis the mapping function value at the ith iteration; γ is the chaos coefficient; X i+1 is the mapping function value at the (i+1)th iteration.

[0101] The meerkat fitness value is calculated, the optimal meerkat fitness value is obtained, and their food sources are determined.

[0102] The excavation phase; in the excavation phase, the meerkat performs an action similar to the shape of a heart line. The heart-shaped movement can be simulated by formula (4):

[0103] First, define the intensity I and the update density factor. The intensity and the concentration of the prey are related to the distance between the meerkat and the prey. I is the intensity of the smell of the prey, the greater the olfactory intensity, the faster the meerkat moves, and vice versa. Specifically as shown in formula (2):

[0104]

[0105] Where s is the source intensity or concentration intensity; d i represents the distance between the prey and the current meerkat individual, and r2 is a random number in (0, 1).

[0106] The density factor ensures a smooth transition from exploration to exploitation. The decreasing factor will decrease with the increase of the number of iterations, and the method in formula (3) is used for updating to reduce the uncertainty brought by the change of time in the simulation of the meerkat foraging process.

[0107]

[0108] Where T max is the maximum number of iterations; C≥1 (generally default to 2).

[0109] Formula (4) simulates the action of the meerkat in the excavation phase:

[0110] x new =x p +F×β×I×x p +F×r3×α×d i ×|cos(2πr4)×[1-cos(2πr5)]| (4)

[0111] Where x p is the global optimal position in the current state; β≥1 (generally default to 6) is the ability of the meerkat to obtain food, d i represents the distance between the prey and the current meerkat individual, r3, r4, r5 are three random numbers in (0, 1), and F is a flag to change the search direction, which is determined by formula (5):

[0112]

[0113] In the excavation phase, meerkats rely heavily on the intensity of the prey's scent, the distance from the prey, and the time-varying search factor α.

[0114] Honey harvesting phase; the situation where meerkats follow honeyguides to the beehive can be simulated using equation (6):

[0115] x new =x p +F×r7×α×d i (6)

[0116] where x new is the updated meerkat individual position; x p is the prey position; F and α are determined by equations (5) and (3), respectively; r7 is a random number in (0, 1). From equation (6), it can be seen that according to the distance information d i , the meerkat searches around the prey position x p .

[0117] In selecting the prediction model, the present application considers a variety of models, including traditional machine learning models such as support vector machines (SVM), random forests (RF), and deep learning models such as convolutional neural networks (CNN), recurrent neural networks (RNN), etc. Finally, the present application chooses the Bi-LSTM model, which has a significant advantage in processing time series data, can effectively capture time dependence, maintain the integrity of information and reduce information loss. Bi-LSTM combines two LSTM networks in forward and reverse directions, which can simultaneously use past and future information, significantly improving the accuracy of prediction.

[0118] In this application, we use the Tent map to improve the THBA algorithm of meerkat initialization population for model optimization. In order to prove the advantages of THBA algorithm compared with other algorithms, we carried out the following comparative cases and experimental data analysis:

[0119] In the field of central air conditioning load forecasting, we adopted the THBA algorithm to optimize the Bi-LSTM model, and constructed the THBA-Bi-LSTM model. Through comparative experiments with GA-Bi-LSTM, PSO-Bi-LSTM and HBA-Bi-LSTM models, the THBA-Bi-LSTM model showed significant advantages in prediction accuracy. The specific experimental data showed that the root mean square error (RMSE) of the THBA-Bi-LSTM model was 8.5, and the mean absolute error (MAE) was 5.2, which were better than the GA-LSTM (RMSE was 11.6, MAE was 7.2), PSO-LSTM (RMSE was 10.5, MAE was 7.1) and HBA-Bi-LSTM (RMSE was 10.0, MAE was 7.0) models, which proved the effectiveness and superiority of the THBA algorithm in optimizing the Bi-LSTM model.

[0120] The optimization ability of the THBA algorithm is due to its improved initialization population strategy through Tent mapping, which significantly improves the global search ability and convergence speed of the algorithm. In the parameter optimization process of the Bi-LSTM model, the THBA algorithm can more effectively explore the solution space and find better parameter combinations, thereby significantly improving the prediction accuracy of the model. In an experiment using the past year's central air conditioning system operation data, the coefficient of determination R 2 As the model prediction accuracy, the prediction accuracy of the THBA-Bi-LSTM model on the test set reached 93%, while the prediction accuracy of the GA-Bi-LSTM, PSO-Bi-LSTM and HBA-Bi-LSTM models was 88%, 90% and 91% respectively, so the prediction accuracy of the THBA-Bi-LSTM model was increased by 5%, 3% and 2% respectively compared with the GA-LSTM and PSO-LSTM models. This shows that the THBA algorithm can more effectively explore the solution space and find better parameter combinations when optimizing the Bi-LSTM model parameters, thereby significantly improving the prediction accuracy of the model.

[0121] In practical applications, the high prediction accuracy of the THBA-Bi-LSTM model provides strong data support for the energy efficiency optimization of central air conditioning systems. By accurately predicting the load demand, the central air conditioning system can be more effectively scheduled and controlled, achieving energy saving and cost reduction. These achievements not only theoretically verify the advantages of the THBA algorithm, but also demonstrate its positive effects in the field of central air conditioning load forecasting in practical applications, providing new optimization tools and methods for this field.

[0122] Therefore, by applying the THBA algorithm to the optimization of the Bi-LSTM model, we not only verify the wide applicability of the THBA algorithm in different fields, but also demonstrate its great potential in practical engineering applications. The high prediction accuracy of the THBA-Bi-LSTM model provides strong data support for the energy efficiency optimization of central air conditioning systems. By accurately predicting load demand, central air conditioning systems can be more effectively scheduled and controlled, achieving energy conservation and cost reduction. These achievements not only theoretically verify the advantages of the THBA algorithm, but also demonstrate its positive effects in the field of central air conditioning aggregated load prediction in practical applications.

[0123] In summary, through the analysis of comparative cases and experimental data, we can conclude that the THBA algorithm has obvious advantages over traditional GA, PSO, and HBA algorithms in terms of global search ability, convergence speed, and prediction effect improvement. These advantages are not only theoretically proven, but also verified in practical applications, providing new optimization tools for technical personnel in this field.

[0124] In summary, by selecting the Bi-LSTM model and THBA optimization algorithm, this application not only performs well in handling time series data with long-term dependencies, but also exhibits stronger search ability and faster convergence speed in the optimization process, providing an efficient and accurate solution for aggregated load prediction of central air conditioning systems.

[0125] The process of optimizing the parameters of Bi-LSTM using the THBA algorithm includes:

[0126] The central air conditioning aggregated power data is used as the input feature variable, and the initial parameters to be optimized are set and given initial values. The parameters to be optimized are the learning rate, the number of neurons in the first hidden layer, the number of neurons in the second hidden layer, and the regularization coefficient.

[0127] Specifically, in HBA, the number of meerkat individuals is set to 20, and the maximum number of iterations is set to 50. The search range of the learning rate of the optimized Bi-LSTM neural network is set to [0.001, 0.01], the search range of the number of nodes in the two hidden layers is set to [10, 80], and the search range of the regularization coefficient is set to [0.0001, 0.001]. To prevent overfitting, the batch size is set to 64, the number of training times is set to 500, and the learning rate drop factor is set to 0.1.

[0128] During the training process, the root mean square error (MSE) between the predicted data and the actual data is used as the fitness function.

[0129] Run the improved THBA algorithm to search and optimize Bi-LSTM, and adjust the parameters to obtain the optimal solution. Specifically, the root mean square error (MSE) calculation formula is as follows:

[0130]

[0131] wherein y t is the actual aggregated load value at time t, is the predicted aggregated load value at time t, and N is the number of samples;

[0132] The optimized Bi-LSTM model is formed by THBA-Bi-LSTM algorithm. Further, the specific steps of the Bi-LSTM algorithm include:

[0133] LSTM is to solve the problem of long-term dependence of sequence by adding memory cells to avoid gradient loss. Its key structure is three "gates", forget gate, input gate and output gate. The role of the forget gate is to select to retain or forget the cell state, the input gate selectively records new information in the cell state, and the output gate is used to deliver the current time t control new cell state c t How much can be output is delivered to LSTM as output value h t . The formulas of each part of LSTM are as follows:

[0134] The calculation formula of the forget gate is as follows:

[0135] f t = σ(W f [h t-1 ,x t ]+b f (8)

[0136] In the formula, W f is the weight matrix of the forget gate; b f is the bias variable of the forget gate, and σ is the activation function sigmoid.

[0137] After the forget gate, there is an input gate, and the input gate also has a weight matrix W i and an input gate bias value b i . The calculation formula is as follows:

[0138] i t = σ(W i [h t-1 ,x t ]+b i (9)

[0139] After determining the forget gate and the input gate, the new cell state r' t of the current input needs to be described, and the calculation formula is as follows:

[0140] r' t = tanh(W r [ht-1 ,x t ]+b r (10)

[0141] Current unit state r t , the calculation method is to rely on the current time forget gate f t And the unit state r t-1 The last time multiplied by the current unit state r' t Multiply the current input i t The result is added and calculated as follows:

[0142] r t =f t ☉r t-1 +i t ☉r' t (11)

[0143] The output gate of the LSTM is calculated as follows:

[0144] o t =σ(W o [h t-1 ,x t ]+b o ) (12)LSTM output gate calculation expression, and the final output of LSTM is related to the output gate and the unit state, and the calculation formula is as follows:

[0145] h t =o t ☉tanh(r t ) (13)

[0146] Bi-LSTM is to pass the signal from the input layer to the network layer through forward LSTM to obtain an output value, which affects the learning rate and output function. At the same time, the output value is calculated through the reverse LSTM, which will determine the connection weight of the input node in Bi-LSTM, so as to form a new set of output values. The calculation process is as follows:

[0147]

[0148] In the formula, x t Is the input at t time, LSTM is a neural unit, Is the output of the backward hidden layer at each time, Is the output of the forward hidden layer at each time;

[0149] If the MSE reaches the minimum, directly use the parameters at this time to construct the Bi-LSTM model to predict the central air conditioning aggregate power, otherwise proceed to the next step;

[0150] If the MSE does not reach the minimum, it is determined whether the maximum number of iterations is reached, and if the maximum number of iterations is reached, the obtained parameters are used to construct the THBA-Bi-LSTM model to realize the aggregated load prediction of the central air conditioner.

[0151] The process of inputting the central air conditioner sample data into the pre-established THBA-Bi-LSTM model and outputting the aggregated load prediction result of the central air conditioner includes:

[0152] Data preprocessing: normalize the input data to the interval [0, 1] to improve the training efficiency of the model, and the normalization formula is as follows:

[0153]

[0154] wherein x is the original data, x' is the normalized data, x min is the minimum value in the original data, x max is the maximum value in the original data, and the original data is the feature data related to the aggregated load.

[0155] The Bi-LSTM model structure, specifically, the input layer: receives input data such as temperature, humidity, and power consumption.

[0156] The Bi-LSTM layer: contains a forward LSTM layer and a reverse LSTM layer, which respectively process the forward and backward information of the sequence data. The forward LSTM layer calculates the forget gate, input gate, new cell state, cell state update, and output gate, and the reverse LSTM layer is similar to the forward LSTM layer, but processes the reverse sequence of the input data.

[0157] The aggregated output layer: outputs the predicted aggregated load value, that is, the outputs of the forward and reverse LSTM layers are combined, and the final prediction result is output through a fully connected layer, and the aggregated load prediction follows the formula:

[0158]

[0159] wherein W y is the output layer weight matrix, b y is the output layer bias vector, is the hidden state of the forward LSTM, is the hidden state of the reverse LSTM, is the aggregated load prediction output value at time t.

[0160] The calculation process of the forward LSTM layer and the reverse LSTM layer in the Bi-LSTM model is described in formulas (7)-(15).

[0161] To verify the performance of the THBA-Bi-LSTM model, we conducted detailed experiments for comparison. The experiments used a year's worth of central air conditioning operation data from a certain commercial building, including temperature, humidity, and power consumption. The selected comparison models included traditional Bi-LSTM, support vector machine (SVM), random forest (RF), GA-LSTM, PSO-LSTM, and HBA-LSTM. The experimental results are shown in the following table:

[0162] Table 1: Aggregated load prediction results of different prediction methods

[0163]

[0164] From the above table, it can be seen that the THBA-Bi-LSTM model outperforms other methods in all evaluation indicators. Specifically, compared with the traditional Bi-LSTM, the RMSE is reduced by 29.2%, the MAE is reduced by 30.7%, and the R 2 is increased by 9.4%. This indicates that the THBA-Bi-LSTM model has significantly improved prediction accuracy.

[0165] The THBA-Bi-LSTM method significantly outperforms other traditional methods in prediction performance. Specifically, compared with the traditional Bi-LSTM, the THBA-Bi-LSTM reduces the RMSE by 29.2%, the MAE by 30.7%, and the R 2 is increased by 9.4%; compared with SVM, the RMSE is reduced by 43.3%, the MAE is reduced by 42.2%, and the R 2 is increased by 19.2%; compared with random forest (RF), the RMSE is reduced by 39.3%, the MAE is reduced by 35.3%, and the R 2 is increased by 16.2%. These results fully demonstrate the significant advantages of the THBA-Bi-LSTM method in improving prediction accuracy and stability.

[0166] These data indicate that the THBA-Bi-LSTM model has significant advantages in predicting the aggregated load of central air conditioning, and can provide more accurate prediction results, thereby helping to more effectively manage and control the load of the central air conditioning system.

[0167] To verify the effectiveness of the THBA-Bi-LSTM model in predicting the aggregated load of the central air conditioning system, we conducted experiments on a central air conditioning system in a certain commercial building. The system contains 3000 air conditioners, and we collected a year's worth of operation data, including hourly operation data of temperature, humidity, and power consumption. Through normalization, feature extraction, and division of the data, we used Latin hypercube sampling to extract 10% of the data as the training set, and the rest as the test set, to ensure the efficiency and generalization ability of the model training.

[0168] In terms of experimental setup, we selected the THBA-Bi-LSTM model for aggregate load prediction and compared it with the traditional LSTM model and the Bi-LSTM model. Evaluation metrics included root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R²). 2 Experimental results show that the THBA-Bi-LSTM model outperforms the traditional model on all evaluation metrics, with RMSE of 8.5, MAE of 5.2, and R0. 2 The accuracy reached 0.93, indicating a significant improvement in its prediction accuracy.

[0169] To further verify the model's practical application effectiveness, we selected three typical scenarios for in-depth analysis: high temperatures in summer, low temperatures in winter, and the transitional season between spring and autumn. During the high-temperature period of summer (July 1st to July 31st), the model's RMSE was 8.0, MAE was 4.8, and R... 2 Up to 0.94; during the cold winter period (January 1st to January 31st), RMSE was 9.0, MAE was 5.6, and R... 2 The RMSE was 0.92; during the spring and autumn transition seasons (March 1st to March 31st and October 1st to October 31st), the RMSE was 8.5, the MAE was 5.2, and the R... 2 The value is 0.93. These results consistently demonstrate that the THBA-Bi-LSTM model maintains high prediction accuracy under both extreme weather and normal conditions.

[0170] In summary, through detailed data collection, processing, and comparative experiments, this study not only demonstrates the superior performance of the THBA-Bi-LSTM model under different climatic conditions, but also provides strong technical support for the energy management and optimization of central air conditioning systems, showcasing its great potential in practical applications.

[0171] Specifically, the present invention will be further illustrated below through embodiments:

[0172] Example 2: Second aspect, such as Figure 3 As shown, in order to achieve the above objectives, the present invention discloses a central air conditioning aggregated load forecasting system, comprising:

[0173] Data processing module 11 is used to receive historical operating data of central air conditioning and to randomly and uniformly sample the historical operating data of central air conditioning using the Latin hypercube sampling (LHS) method to obtain central air conditioning sample data. The historical operating data of central air conditioning includes temperature, humidity and power consumption.

[0174] The load prediction module 12 is configured to input the central air conditioner sample data into a pre-established THBA-Bi-LSTM model, and output a central air conditioner aggregated load prediction result, wherein the pre-established THBA-Bi-LSTM model is obtained by optimizing parameters of Bi-LSTM using a THBA algorithm.

[0175] Based on the same inventive concept, the present application further provides a computer device, comprising one or more processors, and a memory for storing one or more computer programs; the program comprises program instructions, and the processor is configured to execute the program instructions stored in the memory. The processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc., which are the computing core and control core of the terminal, and are configured to implement one or more instructions, and are specifically configured to load and execute one or more instructions in the computer storage medium to implement the above method.

[0176] It needs to be further explained that, based on the same inventive concept, the present application further provides a computer storage medium, which stores a computer program, and the computer program is executed by the processor to perform the above method. The storage medium can adopt any combination of one or more computer readable media. The computer readable medium can be a computer readable signal medium or a computer readable storage medium. The computer readable storage medium can be, but is not limited to, an electrical, magnetic, optical, electrical, magnetic, infrared, or semiconductor system, device or component, or any combination of the above. More specific examples (non-exhaustive list) of the computer readable storage medium include: electrical connections with one or more conductive wires, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above. In the present application, the computer readable storage medium can be any tangible medium containing or storing a program, which can be used by or in combination with an instruction execution system, device or component.

[0177] In the description of the specification, the description of the terms "one embodiment", "an example", "a specific example" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present disclosure. In the specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Also, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in an appropriate manner.

[0178] The basic principles, main features and advantages of the present disclosure are shown and described above. Those skilled in the art should understand that the present disclosure is not limited by the above embodiments, and the above embodiments and descriptions in the specification are only to illustrate the principles of the present disclosure. Without departing from the spirit and scope of the present disclosure, various changes and improvements can be made to the present disclosure, and these changes and improvements all fall within the scope of the claimed present disclosure.

Claims

1. A central air conditioning aggregated load prediction method, characterized in that, The method comprises the following steps: Receiving central air conditioner historical operation data, using Latin hypercube sampling (LHS) to randomly and uniformly sample the central air conditioner historical operation data to obtain central air conditioner sample data, wherein the central air conditioner historical operation data comprises temperature, humidity and power consumption; The LHS simulates the change rule of the air conditioner load power and aggregates the air conditioner load; The central air conditioner sample data is input into a pre-established THBA-Bi-LSTM model to output central air conditioner aggregated load prediction results, wherein the pre-established THBA-Bi-LSTM model is obtained by optimizing the parameters of the Bi-LSTM using the THBA algorithm; The THBA algorithm is obtained by introducing a Tent chaotic mapping to improve the initialization of the meerkat algorithm population, and the THBA algorithm comprises three stages, namely, an initialization stage, a digging stage and a honey collecting stage; The process of optimizing the parameters of the Bi-LSTM using the THBA algorithm comprises: The central air conditioner aggregated power data is used as an input characteristic variable, the initial parameters to be optimized are set and the initial values are given; the parameters to be optimized are the learning rate, the number of neurons in the first hidden layer, the number of neurons in the second hidden layer and the regularization coefficient; In the training process, the root mean square error (MSE) of the prediction data and the actual data is used as the fitness function; the MSE calculation formula is as follows: wherein y t is the actual aggregated load value at time t, is the predicted aggregated load value at time t, and N is the number of samples. The improved THBA algorithm is run to search and optimize the Bi-LSTM, and the parameters are adjusted to obtain the optimal solution.

2. The central air conditioning aggregate load prediction method according to claim 1, characterized in that, The process of using the LHS to randomly and uniformly sample the central air conditioner historical operation data comprises the following steps: Let random variables X1, X2, …, X N cumulative probability distribution function of Y N = F(X N ), first determine the sampling size N, and divide the probability distribution function into N parts. In each interval, take one random sample, Y i = (r i -i-1) / N, where r i is a random number from [0,1]; Y i Substitute into Y N =F(X) N The inverse function F -1 In the above, find the sampled value X in interval i. i ; Repeat the above steps N times to generate N variables subject to the F distribution function.

3. The method of claim 1, wherein, The process of the initialization stage comprises: Set the initialization parameters, including the population size N, the maximum number of iterations T max , generate the initial population of meerkat with chaotic characteristics by using Tent mapping of formula (1) wherein X i is the mapping function value at the i-th iteration; γ is the chaos coefficient; X i+1 is the mapping function value at the i+1-th iteration; Calculate the meerkat fitness value to obtain the optimal meerkat fitness value and determine the food source; In the digging stage, the meerkat performs a heart-shaped linear motion, which is simulated by formula (4): Define the intensity I and the update density factor, I is the smell intensity of the prey, as shown in formula (2): where s is the source intensity or concentration intensity; d i represents the distance between the prey and the current meerkat individual, r2is a random number within (0, 1); The density factor is updated using the method in formula (3): wherein T max is the maximum number of iterations; C ≥ 1; Formula (4) simulates the action of the meerkat in the digging stage: x new = x p + F x β x I x x p + F x r3 x a x d i x |cos(2πr4) x [1 - cos(2πr5)]| (4) where x p is the global optimal position in the current state; β ≥ 1 is the ability of the meerkat to obtain food, d i represents the distance between the prey and the current meerkat individual, r3, r4, r5 are three random numbers in (0, 1), and F is a flag for changing the search direction, which is determined using formula (5): The meerkat reaches the beehive following the honey guide bird, which is simulated by formula (6): x new = x p + F x r7 x a x d i (6) where x new is the updated meerkat individual position; x p is the prey position; F and a are determined by equation (5) and equation (3), respectively; r7 is a random number within (0, 1), and from equation (6), the meerkat searches around the prey position x i according to the distance information d p .

4. The method of claim 1, wherein, The Bi-LSTM comprises a forgetting gate, an input gate and an output gate, wherein the calculation formula of the forgetting gate is as follows: f t = σ(W f [h t-1 ,x t ]+b f (8) In the formula, W f is the weight matrix of the forget gate; b f is the bias variable of the forget gate, and σ is the activation function sigmoid. After the forget gate is the input gate, and there is also a weight matrix W in the input gate i and the bias value b of the input gate i The calculation formula is as follows: i t = σ(W i [ h t-1 , x t ] + b i (9) current input new cell state r' t The calculation formula is as follows: r' t = tanh(W r [h t-1 , x t ] + b r (10) Current cell state r t The calculation method relies on the forgetting gate f at the current time. t The cell state r at the previous time step t-1 After multiplication, the result is compared with the current cell state r'. t Multiply by the current time input i t The results are summed, and the calculation formula is as follows: r t = f t ⊙ r t-1 + i t ⊙ r t (11) The output gate of the LSTM is calculated as shown in the following formula: o t = σ(W o [ h t-1 , x t ] + b o ) (12) The output gate of the LSTM is calculated as shown in the following formula: h t = o t ⊙ tanh(r t ) (13).

5. The method of claim 4, wherein, The Bi-LSTM is a signal transmitted from the input layer to the network layer, and the output value is calculated by the forward LSTM and the output value is calculated by the reverse LSTM, and the calculation process is as follows: In the formula, x t is the input at time t, LSTM is a neural unit, is the output of the backward hidden layer at each time, is the output of the forward hidden layer at each time; If the MSE reaches the minimum, the Bi-LSTM model is directly constructed with the parameters to predict the central air conditioner aggregated power, otherwise the next step is performed; If the MSE does not reach the minimum, it is determined whether the maximum iteration number is reached, and if the maximum iteration number is reached, the obtained parameters are used to construct the THBA-Bi-LSTM model to realize the aggregated load prediction of the central air conditioner.

6. The method of claim 1, wherein, The process of inputting the central air conditioner sample data into the pre-established THBA-Bi-LSTM model to output the aggregated load prediction result of the central air conditioner includes: Data preprocessing: normalize the input data to the [0, 1] interval, and the normalization formula is as follows: wherein x is the original data, x' is the normalized data, x min is the minimum value in the original data, x max is the maximum value in the original data, wherein the original data is the aggregated load-related feature data; The Bi-LSTM model structure includes: an input layer that receives input data, including temperature, humidity, and power consumption; The Bi-LSTM layer includes a forward LSTM layer and a reverse LSTM layer, which process the forward and backward information of the sequence data, respectively, wherein the forward LSTM layer calculates the forget gate, the input gate, the new cell state, the cell state update, and the output gate, and the reverse LSTM layer processes the reverse sequence of the input data; The aggregated output layer outputs the predicted aggregated load value, that is, the outputs of the forward and reverse LSTM layers are combined, and the final prediction result is output through a fully connected layer, and the aggregated load prediction follows the formula: where W y is the output layer weight matrix, b y is the output layer bias vector, is the hidden state of the forward LSTM, is the hidden state of the backward LSTM, is the aggregated load prediction output value at time t.

7. A central air conditioning aggregate load prediction system characterized by, It includes: The data processing module is used for receiving central air conditioner historical operation data, and using Latin hypercube sampling LHS method to randomly and uniformly extract samples from the central air conditioner historical operation data to obtain central air conditioner sample data, wherein the central air conditioner historical operation data includes temperature, humidity and power consumption; The Latin hypercube sampling LHS method simulates the change rule of each air conditioner load power, and aggregates and superimposes the air conditioner load; The load prediction module is used for inputting the central air conditioner sample data into the pre-established THBA-Bi-LSTM model to output the aggregated load prediction result of the central air conditioner, wherein the pre-established THBA-Bi-LSTM model is obtained by optimizing the parameters of Bi-LSTM using THBA algorithm; The THBA algorithm is obtained by introducing Tent chaotic mapping to improve the initialization population of the honey badger algorithm, and the THBA algorithm includes three stages: initialization stage, excavation stage and honey collecting stage; The process of optimizing the parameters of Bi-LSTM using THBA algorithm includes: The aggregated power data of the central air conditioner is used as the input feature variable, the initial parameters to be optimized are set and the initial values are given; the parameters to be optimized are learning rate, first layer hidden layer neuron number, second layer hidden layer neuron number and regularization coefficient; During the training process, the root mean square error MSE of the prediction data and the actual data is used as the fitness function; the root mean square error MSE calculation formula is as follows: wherein y t is the actual aggregated load value at time t, is the predicted aggregated load value at time t, and N is the number of samples. Run the improved THBA algorithm to search and optimize Bi-LSTM, and adjust the parameters to obtain the optimal solution.

Images