Air conditioning energy consumption prediction method, device, equipment and storage medium

Abstract

The present application proposes an air conditioning energy consumption prediction method, device, equipment and storage medium. The air conditioning energy consumption prediction method obtains the operating data of the air conditioner, performs sequence processing on the operating data, determines the target sequence, performs moving weighted calculation on the target sequence through a preset algorithm, and determines a weighted moving mean, which is conducive to capturing the changing law of the target sequence, thereby improving the prediction accuracy of the final target energy consumption; predicting the target energy consumption according to the weighted moving mean, the energy consumption prediction model and the recursive strategy, and predicting the target energy consumption based on the recursive strategy, which is conducive to realizing multi-point prediction and improving the applicability of the prediction; and, it is also applicable to the energy efficiency prediction of air-cooled air conditioners and water-cooled air conditioners, and has a wide range of applications.

Description

Technical Field

[0001] The present application relates to the field of air conditioning, and in particular to an air conditioning energy consumption prediction method, device, equipment and storage medium. Background Art

[0002] In recent years, more and more attention and investment have been paid to energy conservation and emission reduction, which is not only significant in reducing economic costs, but also in promoting the improvement of the ecological environment. At present, the annual energy consumption of IDC computer rooms is close to 2% of my country's total annual energy consumption, which is a veritable energy consumer and a key area of ​​energy conservation and emission reduction in the telecommunications industry. In the energy conservation and emission reduction tasks of IDC computer rooms, reducing the redundancy of air conditioning cooling capacity is a key means to improve the overall energy efficiency of the computer room; regardless of whether the cooling method of the air conditioning equipment is air cooling or water cooling, real-time prediction of air conditioning energy consumption is an important prerequisite for realizing intelligent air conditioning control and reducing cooling capacity redundancy.

[0003] At present, the methods for air conditioning energy consumption prediction can be mainly divided into three categories: the first is the traditional time series method represented by ARIMA; the second is the time series method based on deep learning represented by RNN; the third is the machine learning method represented by GBDT. However, due to some characteristics of air conditioning energy consumption prediction problems, such as time series attributes, complex change patterns, and many influencing factors, these three types of methods cannot obtain generalized and highly applicable energy consumption predictions, and the existing GBDT method can only achieve single-point predictions, which has poor applicability. Summary of the invention

[0004] The embodiments of the present application provide an air conditioning energy consumption prediction method, device, equipment and storage medium to solve at least one problem existing in the related art. The technical solution is as follows:

[0005] In a first aspect, an embodiment of the present application provides a method for predicting air conditioning energy consumption, comprising:

[0006] Get the operating data of the air conditioner;

[0007] Performing sequence processing on the operation data to determine a target sequence;

[0008] Performing a moving weighted calculation on the target sequence using a preset algorithm to determine a weighted moving mean;

[0009] The target energy consumption is predicted according to the weighted moving average, the energy consumption prediction model and the recursive strategy.

[0010] In one embodiment, performing sequence processing on the operating data to determine the target sequence includes:

[0011] Converting the operation data into a regular sequence according to a specified time interval;

[0012] Calculating and determining the autocorrelation coefficient of the regular sequence according to different time steps;

[0013] A hypothesis test is performed based on the autocorrelation coefficient to determine the target sequence.

[0014] In one embodiment, performing hypothesis testing according to the autocorrelation coefficient to determine the target sequence includes:

[0015] Determine the corresponding p value from the probability density function table of the t test according to each of the autocorrelation coefficients;

[0016] Determine the maximum autocorrelation coefficient corresponding to the p-value that is less than the specified significance level, and use the time step corresponding to the maximum autocorrelation coefficient as the target time step;

[0017] A target sequence is determined according to the target time step and the regular sequence.

[0018] In one embodiment, performing moving weighted calculation on the target sequence by a preset algorithm to determine the weighted moving mean includes:

[0019] Determining a moving weight corresponding to each data value in the target sequence;

[0020] Performing weighted calculation according to the moving weight and the data value to obtain a weighted moving mean;

[0021] The moving weights corresponding to the data values ​​gradually increase with the time sequence of the data values.

[0022] In one embodiment, predicting the target energy consumption according to the weighted moving mean, the energy consumption prediction model and the recursive strategy includes:

[0023] Determining predicted energy consumption according to the weighted moving mean and the energy consumption prediction model;

[0024] Returning to the step of obtaining the operating data of the air conditioner to determine a new predicted energy consumption, until all predicted energy consumptions for the specified observation time period are obtained;

[0025] Sum up all predicted energy consumptions to get the target energy consumption.

[0026] In one implementation, determining the predicted energy consumption according to the weighted moving mean and the energy consumption prediction model includes:

[0027] Inputting the weighted moving mean into the energy consumption prediction model, and performing inverse conversion processing on the output of the energy consumption prediction model to obtain the predicted energy consumption;

[0028] or,

[0029] The weighted moving average is input into the energy consumption prediction model, and the output value of the energy consumption prediction model is used as the predicted energy consumption.

[0030] In one embodiment, the energy consumption prediction model is trained by the following steps:

[0031] Get training data;

[0032] The energy consumption prediction network is trained by the GBDT algorithm and the training data, and a hyperparameter combination is generated by the Latin hypercube sampling method during the training process, and the performance index of the energy consumption prediction network is obtained by participating in the training through the hyperparameter combination;

[0033] The energy consumption prediction model is determined based on the hyperparameter combination corresponding to the best performance indicator.

[0034] In a second aspect, an embodiment of the present application provides an air conditioning energy consumption prediction device, comprising:

[0035] An acquisition module is used to acquire the operation data of the air conditioner;

[0036] A sequence module, used for performing sequence processing on the operation data to determine a target sequence;

[0037] A calculation module, used to perform moving weighted calculation on the target sequence through a preset algorithm to determine a weighted moving mean;

[0038] The prediction module is used to predict the target energy consumption according to the weighted moving mean, the energy consumption prediction model and the recursive strategy.

[0039] In one embodiment, the air conditioning energy consumption prediction device also includes a training module, which is used to: obtain training data; train the energy consumption prediction network through the GBDT algorithm and the training data, and generate hyperparameter combinations through the Latin hypercube sampling method during the training process, and participate in the training through the hyperparameter combinations to obtain the performance indicators of the energy consumption prediction network; determine the energy consumption prediction model according to the hyperparameter combination corresponding to the best performance indicator.

[0040] In a third aspect, an embodiment of the present application provides an electronic device, comprising: a processor and a memory, wherein the memory stores instructions, and the instructions are loaded and executed by the processor to implement a method in any one of the above-mentioned embodiments.

[0041] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed, the method in any one of the above-mentioned embodiments is implemented.

[0042] The beneficial effects of the above technical solution include at least:

[0043] By acquiring the operating data of the air conditioner, performing sequence processing on the operating data, determining the target sequence, and performing moving weighted calculation on the target sequence through a preset algorithm, and determining the weighted moving mean, it is helpful to capture the changing law of the target sequence, thereby improving the prediction accuracy of the final target energy consumption; predicting the target energy consumption according to the weighted moving mean, the energy consumption prediction model and the recursive strategy, and predicting the target energy consumption based on the recursive strategy, it is helpful to achieve multi-point prediction and improve the applicability of the prediction; and it is also applicable to the energy efficiency prediction of air-cooled air conditioners and water-cooled air conditioners, and has a wide range of applications.

[0044] The above summary is for illustrative purposes only and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments and features described above, further aspects, embodiments and features of the present application will be readily apparent by reference to the accompanying drawings and the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In the accompanying drawings, unless otherwise specified, the same reference numerals throughout the multiple drawings represent the same or similar parts or elements. These drawings are not necessarily drawn to scale. It should be understood that these drawings only depict some embodiments disclosed in the present application and should not be regarded as limiting the scope of the present application.

[0046] Figure 1 This is a schematic diagram of the steps of an air conditioning energy consumption prediction method according to an embodiment of the present application;

[0047] Figure 2 A schematic diagram of a decision tree node according to an embodiment of the present application;

[0048] Figure 3 A schematic diagram of a cross-validation method according to an embodiment of the present application;

[0049] Figure 4 A schematic diagram of a recursive strategy according to an embodiment of the present application;

[0050] Figure 5 A schematic diagram comparing the energy consumption of air conditioners in one embodiment of the present application;

[0051] Figure 6 This is a structural block diagram of an air conditioning energy consumption prediction device according to an embodiment of the present application;

[0052] Figure 7 This is a structural block diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0053] In the following, only some exemplary embodiments are briefly described. As those skilled in the art will appreciate, the described embodiments may be modified in various ways without departing from the spirit or scope of the present application. Therefore, the drawings and descriptions are considered to be exemplary and non-restrictive in nature.

[0054] Glossary:

[0055] ARIMA (Autoregressive Integrated Moving Average): also known as the integrated moving average autoregressive model (moving can also be called sliding), is a time series forecasting and analysis method.

[0056] GBDT (Gradient Boosting Decision Tree): It is a type of machine learning algorithm that uses decision trees as weak learners and obtains strong learners through gradient boosting and integration. It can be used to deal with regression and classification problems.

[0057] RNN (Recurrent Neural Network) is a type of recursive neural network that takes sequence data as input, recursively in the direction of sequence evolution, and all nodes are connected in a chain.

[0058] XGBoost (eXtreme Gradient Boosting): is a software implementation of the GBDT algorithm, with the main advantages of high efficiency, flexibility, and portability.

[0059] Data leakage: The training data set used in the training phase of the machine learning model contains information related to the label, which makes the model's performance on the training set overestimated but the generalization ability on new data is poor.

[0060] Among the related technologies, most traditional time series methods such as ARIMA are statistical parameter methods, which have specific assumptions about the original distribution of data, and are difficult to learn the complex laws in the data. In addition, other independent variables cannot be introduced into the model, so underfitting is prone to occur. Deep learning methods such as RNN require a large amount of high-quality training data as a prerequisite for model prediction accuracy. However, in actual production, there may be many problems such as measurement errors and transmission failures, and it is usually impossible or difficult to obtain sufficient high-quality data. Machine learning methods such as GBDT are not specifically designed to solve time series problems. They cannot directly accept time series data as input, cannot directly capture the autocorrelation of time series, are easily affected by data leakage, and cannot output multi-point predictions. Therefore, they are not directly used for time series prediction. In summary, the above problems often lead to a series of adverse consequences such as low accuracy of air conditioning energy consumption prediction, poor generalization, and high construction and maintenance costs.

[0061] Reference Figure 1 , a flow chart of an air conditioning energy consumption prediction method according to an embodiment of the present application is shown, and the air conditioning energy consumption prediction method may at least include steps S100-S400:

[0062] S100: Acquire operating data of the air conditioner.

[0063] S200: Perform sequence processing on the operation data to determine the target sequence.

[0064] S300: Perform moving weighted calculation on the target sequence using a preset algorithm to determine a weighted moving mean.

[0065] S400: predicting target energy consumption according to the weighted moving average, the energy consumption prediction model and the recursive strategy.

[0066] The air conditioning energy consumption prediction method of the embodiment of the present application can be executed by an electronic control unit, a controller, a processor of a terminal such as a computer, a mobile phone, a tablet, a vehicle-mounted terminal, or by a cloud server.

[0067] The technical solution of the embodiment of the present application obtains the operating data of the air conditioner, performs sequence processing on the operating data, determines the target sequence, performs moving weighted calculation on the target sequence through a preset algorithm, and determines a weighted moving mean, which is conducive to capturing the changing pattern of the target sequence, thereby improving the prediction accuracy of the final target energy consumption; predicting the target energy consumption according to the weighted moving mean, the energy consumption prediction model and the recursive strategy, and predicting the target energy consumption based on the recursive strategy, is conducive to achieving multi-point prediction and improving the applicability of the prediction; and, it is also applicable to the energy efficiency prediction of air-cooled air conditioners and water-cooled air conditioners, and has a wide range of applications.

[0068] In one embodiment, the operating data in step S100 includes but is not limited to return air temperature / humidity and its set value, ambient temperature / humidity and other signal data and its collection time, etc.; in some embodiments, when there are multiple air conditioners, the operating data may also include coding data of the equipment, and the coding data is used to distinguish data of different air conditioners.

[0069] Optionally, the above-mentioned operating data and basic data for training can be collected by a signal acquisition unit. The basic data may include air conditioning energy consumption, the above-mentioned return air temperature / humidity and its set value, ambient temperature / humidity and other signal data and its collection time, etc.

[0070] In the embodiment of the present application, the signal acquisition unit transmits all signals generated by the equipment such as the computer room air conditioner, electric meter, temperature and humidity sensor to the computer room power environment monitoring system through the B interface protocol, and requests and obtains the required data through the REST API provided by the computer room power environment monitoring system. It should be pointed out that if the energy consumption to be predicted is the sum of the air conditioning energy consumption of a single IDC computer room, since a computer room contains multiple air conditioners, when collecting signals such as air conditioning energy consumption, air conditioning return air temperature / humidity and its set value, it is necessary to collect each air conditioning device separately, and each air conditioner is configured with a device code. Then, various data are written into the data storage unit and the stored data are read from the storage unit through the data reading and writing unit. The data to be read or written include: air conditioning equipment code, air conditioning energy consumption, air conditioning return air temperature / humidity and its set value, ambient temperature / humidity and other signals, various signal collection times, program running logs, etc.

[0071] Among them, data reading and writing can be: writing various data collected by the signal acquisition unit and the generated program running logs to the hard disk for persistent storage through the API provided by the data storage unit; reading various signal data or program running logs required for use from the hard disk through the API provided by the data storage unit. In addition, the data storage unit can be used to store various signal data, as well as log information generated by each unit in the process of performing various tasks, including: air conditioning equipment code, air conditioning energy consumption, air conditioning return air temperature / humidity and its set value, ambient temperature / humidity and other signals, various signal collection times and program running logs, etc.

[0072] The data storage unit of the embodiment of the present application interacts with the data read-write unit by providing an API interface. When storing data, the data storage unit must write the data to the hard disk for persistent storage to avoid data loss caused by equipment power failure and other failures. At the same time, data backup and disaster recovery functions can be introduced when necessary. In addition, in order to avoid waste of resources caused by unnecessary read and write operations and improve response speed, the data storage unit will partition the data for storage.

[0073] In one embodiment, step S200 includes steps S210-S230:

[0074] S210 , converting the operating data into a regular sequence according to a specified time interval.

[0075] Optionally, when running data acquisition, data can be generated according to a preset fixed period, but due to factors such as timing errors and data processing delays, the actual acquisition frequency will fluctuate around the preset frequency. Since most time series data processing operations assume that the sequence sampling interval is fixed, the non-fixed sampling interval will cause inconvenience to the processing of the time series, model training and prediction in subsequent steps. Therefore, the data processing unit needs to process the original signal sequence to obtain a time series with regular intervals, i.e., a regular sequence, for use in subsequent steps.

[0076] In order to convert the original sequence (operation data) into a regular sequence, it is necessary to actively extend the intervals between the signal values ​​of various types of signals (i.e., various data types of operation data), and assume that the sampling time of various types of signals is the same, and then use linear interpolation to determine the signal value at each moment, and finally obtain a new regular sequence. If v represents the new sequence (regular sequence), v' represents the original sequence of the original signal, t and t' represent the time points in the regular sequence and the original sequence respectively, and i and j represent the serial numbers of the time points in the regular sequence and the original sequence respectively, then The size of the original sequence depends on the i The two closest moments t′ j , t′ j+1 (t′ j ≤t i <t′ j+1 ) corresponds to the signal value, and their specific relationship is shown in Definition 1:

[0077]

[0078] S220, calculating and determining the autocorrelation coefficient of the regular sequence according to different time steps.

[0079] It should be noted that the energy consumption of air-conditioning equipment and the changes in various signals such as ambient temperature have autocorrelation, and the waveform is relatively complex and periodic, and the GBDT algorithm cannot directly capture the autocorrelation. Therefore, in order to enable the algorithm to learn these laws more directly and efficiently, the embodiment of the present application uses statistical methods to determine the signal change cycle, find the historical time period that is closest to the size of any time period sequence, and use the found historical data to construct features in subsequent steps.

[0080] Specifically, the autocorrelation coefficient is used as a measure of the similarity of sequence changes, and the signal change period is determined by combining the coefficient size and the hypothesis test results: for each moment t of the regularized signal sequence i , respectively calculate the autocorrelation coefficient of the signal value of the regular sequence in a period of time before the moment with multiple different time steps s

[0081] For example, given any time t i and step length s j , defined by the following formula from v t Take two subsequences x from k and k (where t i -s j -k≥1):

[0082]

[0083] Then, at time t i and step length s j The corresponding autocorrelation coefficient of the original sequence It can be calculated according to the following formula: are the average values ​​calculated using the signal values ​​in the corresponding time step s:

[0084]

[0085] S230. Perform hypothesis testing based on the autocorrelation coefficient to determine the target sequence.

[0086] In one implementation, step S230 includes steps S2301-S2303:

[0087] S2301. Determine the corresponding p value from the probability density function table of the t-test according to each autocorrelation coefficient.

[0088] S2302. Determine the maximum autocorrelation coefficient corresponding to the p-value that is less than the specified significance level, and use the time step corresponding to the maximum autocorrelation coefficient as the target time step.

[0089] S2303: Determine the target sequence according to the target time step and the rule sequence.

[0090] It should be noted that since the autocorrelation coefficient follows the Student's t (t test) distribution in statistics, the corresponding p value can be determined from the probability density function table of the corresponding t test according to each autocorrelation coefficient by calculating the autocorrelation coefficient. Then, it is determined whether the p value is less than the significance level α, for example, the significance level α = 0.05 is t iFind the autocorrelation coefficient with p<α=0.05 and the maximum value The maximum autocorrelation coefficient The corresponding time step is taken as the target time step s j ', then we can determine And the target sequence composed of the historical data with the strongest correlation:

[0091] (where t i -ks j ′≥1), in the subsequent process, the moving mean algorithm can be used to further construct features based on these target sequences.

[0092] Among them, The corresponding Student's t value The calculation method is as follows, n represents the subsequence x k or k Sample size:

[0093]

[0094] In one implementation, step S300 includes steps S310-S320:

[0095] S310: Determine the moving weight corresponding to each data value in the target sequence.

[0096] In the embodiment of the present application, the moving weight w corresponding to each data value gradually increases with the time sequence of the data value, that is, the items near the end of the target sequence have a higher weight in the calculation result, and vice versa. Among them, the moving weight w is a feature engineering hyperparameter introduced in this embodiment, and its specific size needs to be determined by cross-validation; given multiple different values, the cross-validation method is used to fit and evaluate the model, and the optimal value is selected according to the evaluation results.

[0097] S320, performing weighted calculation according to the moving weight and the data value to obtain a weighted moving mean.

[0098] In the embodiment of the present application, the weighted moving average The specific calculation formula is:

[0099]

[0100] In one embodiment, the energy consumption prediction model is trained by following steps S1-S3:

[0101] S1. Obtain training data.

[0102] Optionally, the training data may include return air temperature / humidity and its set value, ambient temperature / humidity and other signal data and its collection time, energy consumption value labels (hereinafter referred to as labels), etc. It should be noted that the present application takes multiple air-conditioning devices as an example, and the training data includes the coding data of the air-conditioning devices. For example, the air-conditioning devices are numbered to determine the coding data of non-negative integers, and then the air-conditioning energy consumption of all air-conditioning devices is scaled to the interval of [-0.1, 0.1], and the corresponding coding data is added to obtain the label. If y i Represents the original energy consumption value label, y′ i represents the label of the i-th air conditioner, y i,* Represents the energy consumption sequence and extreme value of the same air-conditioning equipment, * can be min or max, then the process can be expressed as:

[0103]

[0104] Among them, the new label will replace the original energy consumption sequence and be used for model training; at the same time, the air-conditioning equipment code i will be used as a new feature for model training and prediction.

[0105] S2. The energy consumption prediction network is trained using the GBDT algorithm and training data, and hyperparameter combinations are generated during the training process using the Latin hypercube sampling method and the cross-validation method. The hyperparameter combinations are used for training to obtain the performance indicators of the energy consumption prediction network.

[0106] S3. Determine the energy consumption prediction model based on the hyperparameter combination corresponding to the best performance indicator.

[0107] Optionally, the energy consumption prediction network is trained using the GBDT algorithm and training data. When the GBDT algorithm is used to train the model, the root mean square error is used as the loss function. represents the prediction result, y represents the actual observation value, n represents the sample size, then L represents the loss function calculation result which can be expressed as:

[0108]

[0109] Among them, after converting the labels, during the model training process, the GBDT algorithm must divide the air conditioners in the first few layers of nodes in each decision tree so that the subsequent child nodes only have data for one air conditioner ( Figure 2 ). The final model is equivalent to containing multiple sub-models, each of which is optimized for one air conditioner. The principle is that label conversion amplifies the difference between label values ​​of different devices. If the algorithm does not first consider the device difference (corresponding to the new feature i) during the fitting process and divides different devices into different sub-nodes, the loss function result value will be greatly amplified.

[0110] Assume that y′ represents the energy consumption label that the model needs to predict and has been converted to a numerical value, X t represents a series of input features of the model, and t represents the time corresponding to the label value or feature value. Their relationship in the energy consumption prediction model is:

[0111] y′ t+1 =f(X t )

[0112] The time t+1 corresponding to the label is always one moment later than the time t corresponding to the input feature. This means that when organizing the training data set, for each label in the record, the corresponding feature should only be constructed from the data of the previous moment or earlier than the moment corresponding to the label to avoid data leakage problems.

[0113] The present invention uses the GBDT algorithm to construct an energy consumption prediction model according to the above relationship, and after selecting the model performance evaluation index, conducts training and optimizes various hyperparameters.

[0114] At the same time, during the training process, hyperparameter combinations are generated through Latin hypercube sampling and cross-validation methods, and the performance indicators of the energy consumption prediction network are obtained through the hyperparameter combinations. For example, a reasonable sampling interval is specified for each hyperparameter, and then the Latin hypercube sampling method is used to obtain the value combination of multiple hyperparameters in the sample space, and then the cross-validation method based on time series is used to obtain the model performance indicators obtained by training with each hyperparameter combination; finally, the hyperparameter combination corresponding to the model with the best performance indicator is selected as the optimal solution, and these optimal solutions are used to train on the full amount of data to obtain the final model, thereby determining the energy consumption prediction model.

[0115] like Figure 3 As shown, a cross-validation method based on time series is shown. Figure 3 Each row in the figure represents a model training and evaluation process, and each dot represents the data corresponding to a moment in the time series; black dots represent data used to construct features, gray dots represent data used to construct labels, and white dots represent data that do not participate in the current model training and evaluation. Figure 3 The model is trained and evaluated 10 times in this way, and each new process is 1 step longer than the previous one. After completing the 10 processes, the 10 evaluation indicators are averaged to obtain the comprehensive model performance corresponding to the hyperparameter value combination, so as to guide the selection of the final hyperparameter value combination.

[0116] It should be noted that the energy consumption label value used in the fitting model is based on the above y′ iis obtained by transforming the formula, so the predicted value of the i-th air conditioner device output by the model should also be regarded as the result of transformation and needs to be inversely transformed to regard the result as the true energy consumption prediction This inverse transformation process is to subtract the code i of the corresponding air conditioner from the predicted value and then perform anti-normalization. The process is as follows:

[0117]

[0118] where y i,nax and y i,min represent the maximum and minimum values of the original label of the i-th air conditioner device respectively.

[0119] It should be noted that this application adopts a recursive prediction strategy to achieve multi-point prediction:

[0120] Assume that t = 1, 2,..., T represent the time points corresponding to the data included in the training data, and t = T + i (i = 1, 2,..., n) represent the time points corresponding to the labels that the model needs to predict. Then, based on the inverse transformation of the labels, the recursive prediction process adopted by the present invention can be expressed as follows:

[0121] 1. Use the data processing unit to process the training data from t = 1 to t = T + i - 1 to obtain the feature X T+i-1 , and the processing here is to determine the weighted moving average corresponding to the training, and the principle is as described above;

[0122] 2. Input X T+i-1 into the model and obtain the preliminary prediction result

[0123] 3. Inversely transform to obtain

[0124] 4. Add to the original observed data as the new observed data;

[0125] 5. If i < n, then let i = i + 1 and repeat the above steps, otherwise consider the entire recursive prediction process to be completed.

[0126] In one implementation, step S400 includes steps S410 - S430:

[0127] S410. Determine the predicted energy consumption according to the weighted moving average and the energy consumption prediction model.

[0128] Optionally, step S410 may include step S4101 or S4102:

[0129] S4101. Input the weighted moving mean into the energy consumption prediction model, and perform inverse conversion processing on the output of the energy consumption prediction model to obtain predicted energy consumption.

[0130] Optionally, since multiple air conditioners may be configured to cool simultaneously in an actual IDC computer room, the total energy consumption of the air conditioning equipment in a computer room is the sum of the energy consumption of multiple air conditioning equipment. On the one hand, when constructing an energy consumption prediction model, if sub-models are constructed for different air conditioning equipment respectively, the number of models required for maintenance will increase and the maintenance cost will increase. On the other hand, differences in brands, specifications, and depreciation levels of different air conditioning equipment may lead to differences in the laws of energy consumption changes; in the presence of such differences, if the data of all devices are directly used to fit a single model, the effect may not be ideal. Therefore, the embodiment of the present application solves the problem of a single model by converting the energy consumption label values ​​of different air conditioning equipment, affecting the calculation of the loss function, and adding the equipment code to the model features, that is, the energy consumption prediction model is obtained by training the air conditioning energy consumption after the above-mentioned conversion process, so the weighted moving mean is input into the energy consumption prediction model, and after obtaining the output of the energy consumption prediction model, the output needs to be inversely converted to obtain the predicted energy consumption.

[0131] S4102: Input the weighted moving average into the energy consumption prediction model, and use the output value of the energy consumption prediction model as the predicted energy consumption.

[0132] Optionally, if only one air conditioner is actually considered, energy consumption encoding may not be performed, and the energy consumption prediction model may be trained using the air conditioner energy consumption without conversion processing. At this time, the weighted moving mean is input into the energy consumption prediction model, and the output value of the energy consumption prediction model is used as the predicted energy consumption.

[0133] S420, returning to the step of obtaining the operating data of the air conditioner to determine a new predicted energy consumption, until all predicted energy consumptions in the specified observation time period are obtained.

[0134] It should be noted that in existing solutions, such as the existing GBDT model, by default, only one time point can be predicted at a time. In order to enable the model to output predicted values ​​for multiple future time points, this embodiment uses a recursive strategy to call the model to generate the required output. After determining a predicted energy consumption, it returns to the step of obtaining the operating data of the air conditioner, that is, step S100, so as to determine a new predicted energy consumption according to the same principle until all predicted energy consumptions in the specified observation time period are obtained, thereby achieving the purpose of multi-point prediction. For example, Figure 4 Describes the recursive prediction iteration process for the specified time period T-T+5.

[0135] S430: Sum all predicted energy consumptions to obtain target energy consumption.

[0136] For example, when the predicted energy consumption at each time point in each time period is determined to be The final target energy consumption is obtained by summing them up.

[0137] The following is a description of the embodiments of the present application using a specific embodiment in which the cooling method of the air-conditioning equipment is air cooling. It should be noted that in other embodiments, the cooling method of the air-conditioning equipment may also be water-cooled, which does not constitute a specific limitation. The method of the embodiments of the present application is applicable to both air-cooled and water-cooled air-conditioning equipment, and can achieve the required prediction effect and prediction standard:

[0138] Collect operation data, the field information of the operation data can be as shown in Table 1:

[0139] Table 1

[0140]

[0141]

[0142] In this embodiment, the original acquisition frequency of each type of signal is preset to 1 time / minute. In order to obtain a time series with a regular interval, this embodiment resets the interval to 2 minutes, and then uses linear interpolation to determine the signal value at each moment to obtain a new rule. This embodiment performs this processing on all signal sequences, including air conditioning energy consumption, air conditioning return air temperature / humidity and its set value, and ambient temperature / humidity. Taking the air conditioning energy consumption signal as an example, this embodiment compares some of the results before and after processing it. Figure 5 shown.

[0143] This embodiment uses 15 steps (s=[1,2,3,...,15]) when calculating the autocorrelation coefficient of each sequence. Because this embodiment adjusts the sampling interval to 2 minutes, the time intervals corresponding to the above steps are [2,4,6,...,30] (unit: minutes). Since the change period of each sequence in this embodiment is basically within 30 minutes, it can be covered by the maximum value of the interval in most cases.

[0144] This embodiment calculates the autoregressive coefficients for the air conditioning energy consumption, air conditioning return air temperature / humidity, ambient temperature / humidity and other sequences, and uses the results to guide feature engineering. This embodiment does not perform such processing on the air conditioning return air temperature / humidity set value, time, date and related derived features.

[0145] In this embodiment, after obtaining a new derivative sequence by calculating the autocorrelation coefficient method for the air conditioning energy consumption, air conditioning return air temperature / humidity, ambient temperature / humidity and other sequences, the weighted moving average algorithm is used to obtain the weighted moving average corresponding to each moment for the derivative sequence. Taking part of the data of the energy consumption sequence of a certain air conditioning equipment in this embodiment as an example, the weight w=0.25 is taken, and the relationship between the original value of the signal and the weighted moving average is shown in Table 2.

[0146] Table 2

[0147]

[0148]

[0149] The computer room in this embodiment is equipped with three air conditioners that provide cooling at the same time, so the total energy consumption of the air conditioner in the computer room is the sum of the energy consumption of the three air conditioners. This embodiment converts the energy consumption label values ​​of different air conditioners, affects the calculation of the loss function, and adds the device code to the model feature.

[0150] In this embodiment, three air conditioners are numbered in any order to obtain non-negative integer codes for each device; then, the energy consumption values ​​of all air conditioners are scaled to the interval [-0.1, 0.1], and the corresponding codes are added to obtain energy consumption value labels. In this embodiment, the minimum values ​​of the energy consumption labels of the three air conditioners are 0.12, 0.35, and 0.33, respectively, and the maximum values ​​are 1.95, 1.8, and 2.1, respectively. Taking some data as an example, the label conversion results are shown in Table 3:

[0151] Table 3

[0152]

[0153]

[0154] In this embodiment, when training the model, the input features used specifically include: air conditioning equipment code,

[0155] Weighted moving average of energy consumption (data filtered by autocorrelation coefficient), weighted moving average of ambient temperature (data filtered by autocorrelation coefficient), weighted moving average of ambient humidity (data filtered by autocorrelation coefficient), air conditioning return air temperature set value, air conditioning return air humidity set value, difference between ambient temperature and air conditioning return air temperature set value, difference between ambient humidity and air conditioning return air humidity set value, duration of the last change in ambient temperature set value, duration of the last change in ambient humidity set value, size of the last change in ambient temperature set value, size of the last change in ambient humidity set value, date (Monday to Sunday), hour (0:00-23:00).

[0156] This embodiment constructs an energy consumption prediction model by using the GBDT algorithm; specifically, the GBDT software used in this embodiment is implemented as XGBoost. When training the model, this embodiment uses the mean squared error (MSE) as the algorithm loss function and also uses it as an evaluation index of the overall effect of the model. At the same time, the hyperparameter combination optimized when building the model is as follows: feature engineering hyperparameters, weights (w), learning rate (learning_rate), number of trees (n_estimators), tree depth (max_depth), sample ratio (subsample), number of features (colsample_bytree), early stopping rounds (early_stopping_rounds).

[0157] Table 4 lists the hyperparameter combinations obtained by Latin hypercube sampling in this embodiment, and their corresponding cross-validation results.

[0158] Table 4

[0159]

[0160] The results show that the optimal parameter value combination of this embodiment makes the mean square error only 1.62e-02, and its specific values ​​are as follows: weight: 0.25, learning rate: 6.30e-03, number of trees: 1938, tree depth: 15, sample ratio: 0.77, number of features 7, early stopping rounds: 15.

[0161] After determining the optimal values ​​of each hyperparameter through cross-validation, this embodiment uses these parameter values ​​to conduct further training on the full amount of data to obtain the final model.

[0162] This embodiment needs to predict the energy consumption in the next 10 minutes; and since the sampling interval of the time series is set to 2 minutes in this embodiment, the model needs to output the predicted values ​​of 5 consecutive time points in the future. However, the XGBoost model itself only supports the output of the predicted value of a single time point. In order to enable the model to output the predicted values ​​of multiple time points in the future, this embodiment uses a recursive strategy to call the model, generate the required output, and perform a reverse conversion and aggregation to obtain the final result. The predicted label reverse conversion results are shown in Table 5.

[0163] Table 5

[0164]

[0165]

[0166] Among them, after obtaining the inverse conversion results of the predicted labels of each device at each time point, they are summed up by device to obtain the final output of the model. Taking the data shown in Table 5 as an example, the final predicted results of the energy consumption of air-conditioning equipment 1, 2, and 3 are 4.96, 2.1, and 5.56 (unit: kWh) respectively.

[0167] By using the method of the embodiment of the present application, at least the following effects can be achieved:

[0168] 1) The present invention proposes to use the autocorrelation coefficient to guide the feature engineering of the energy consumption prediction model. The machine learning algorithm represented by GBDT in the prior art is not specially designed to solve the time series problem, and cannot capture the autocorrelation characteristics in the time series, which will cause the underfitting of the model; the present invention introduces the autocorrelation coefficient to guide the feature engineering, guides the feature engineering through the autocorrelation coefficient and uses the moving weighted mean algorithm to construct key features, so that the model can capture the autocorrelation characteristics of the time series and learn the complex changes of waveform amplitude, cycle length, trend level, etc., thereby improving the prediction accuracy of the model, and the energy consumption prediction model finally obtained can obtain more accurate prediction results.

[0169] 2) Use the weighted moving average algorithm to construct the key features involved in energy consumption prediction.

[0170] Since the energy consumption of air-conditioning equipment varies in a complex pattern and is affected by many factors, its waveform, cycle length, trend level and other characteristics are constantly changing. In the face of these complex changes, existing methods often cannot give accurate predictions. The present invention proposes a weighted moving average algorithm with a hyperparameter for constructing key features, so that the algorithm can better learn the complex change pattern during the training phase and ultimately output more accurate prediction results.

[0171] 3) Propose a method to guide the algorithm to perform in-depth optimization for specific situations by converting labels and affecting the calculation of loss functions.

[0172] 4) The computer room often contains multiple air-conditioning equipment, and the brands, specifications, depreciation levels, etc. of these equipment may not be consistent. The current common method is to either not distinguish between the equipment and put the data of all equipment into one model for training, which reduces the prediction accuracy; or to train a model for each device separately, resulting in an increase in the number of models and maintenance costs. The method proposed by the invention guides the model to perform special optimization on the equipment, thereby improving the prediction accuracy while reducing the number of models and maintenance costs;

[0173] 5) By blocking the penetration of label information into labels during feature engineering, the overfitting problem caused by data leakage is avoided, thus improving the generalization ability of the model;

[0174] 6) By affecting the calculation of the loss function, the model can be deeply optimized for specific conditions, improving the prediction accuracy;

[0175] 7) Through the recursive strategy, the model can achieve multi-point output of prediction results.

[0176] Reference Figure 6 , shows a structural block diagram of an air conditioning energy consumption prediction device according to an embodiment of the present application, the device may include:

[0177] An acquisition module is used to acquire the operation data of the air conditioner;

[0178] The sequence module is used to perform sequence processing on the running data and determine the target sequence;

[0179] A calculation module, used to perform moving weighted calculation on the target sequence through a preset algorithm to determine a weighted moving mean;

[0180] The prediction module is used to predict the target energy consumption according to the weighted moving mean, the energy consumption prediction model and the recursive strategy.

[0181] The functions of each module in each device in the embodiments of the present application can be found in the corresponding description in the above method, and will not be repeated here.

[0182] Reference Figure 7 , shows a structural block diagram of an electronic device according to an embodiment of the present application, the electronic device includes: a memory 310 and a processor 320, the memory 310 stores instructions that can be run on the processor 320, and the processor 320 loads and executes the instructions to implement the air conditioning energy consumption prediction method in the above embodiment. The number of the memory 310 and the processor 320 can be one or more.

[0183] In one embodiment, the electronic device also includes a communication interface 330 for communicating with external devices and performing data exchange transmission. If the memory 310, the processor 320 and the communication interface 330 are implemented independently, the memory 310, the processor 320 and the communication interface 330 can be interconnected through a bus and complete mutual communication. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 7Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0184] Optionally, in a specific implementation, if the memory 310, the processor 320 and the communication interface 330 are integrated on a chip, the memory 310, the processor 320 and the communication interface 330 can communicate with each other through an internal interface.

[0185] An embodiment of the present application provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the air conditioning energy consumption prediction method provided in the above embodiment.

[0186] An embodiment of the present application also provides a chip, which includes a processor for calling and executing instructions stored in the memory from the memory, so that a communication device equipped with the chip executes the method provided by the embodiment of the present application.

[0187] An embodiment of the present application also provides a chip, including: an input interface, an output interface, a processor and a memory, wherein the input interface, the output interface, the processor and the memory are connected via an internal connection path, and the processor is used to execute the code in the memory. When the code is executed, the processor is used to execute the method provided in the embodiment of the application.

[0188] It should be understood that the processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. It is worth noting that the processor may be a processor supporting the advanced RISC machines (ARM) architecture.

[0189] Further, optionally, the above-mentioned memory may include a read-only memory and a random access memory, and may also include a non-volatile random access memory. The memory may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may include a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may include a random access memory (RAM), which is used as an external cache. By way of example but not limitation, many forms of RAM are available. For example, static RAM (SRAM), dynamic random access memory (DRAM), synchronous DRAM (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM) and direct rambus RAM (DR RAM).

[0190] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function according to the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium.

[0191] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification and the features of different embodiments or examples, unless they are contradictory.

[0192] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0193] Any process or method description in the flow chart or otherwise described herein can be understood to represent a module, fragment or portion of a code including one or more executable instructions for implementing the steps of a specific logical function or process. And the scope of the preferred embodiment of the present application includes other implementations, in which the functions may not be performed in the order shown or discussed, including in a substantially simultaneous manner or in a reverse order according to the functions involved.

[0194] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, which can be embodied in any computer-readable medium for use by an instruction execution system, apparatus or device (such as a computer-based system, a system including a processor or other system that can fetch instructions from an instruction execution system, apparatus or device and execute instructions), or used in combination with these instruction execution systems, apparatuses or devices.

[0195] It should be understood that the various parts of the present application can be implemented with hardware, software, firmware or a combination thereof. In the above embodiments, multiple steps or methods can be implemented with software or firmware stored in a memory and executed by a suitable instruction execution system. All or part of the steps of the above embodiment method can be completed by instructing the relevant hardware through a program, which can be stored in a computer-readable storage medium, and when the program is executed, it includes one of the steps of the method embodiment or a combination thereof.

[0196] In addition, each functional unit in each embodiment of the present application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into one module. The above-mentioned integrated module can be implemented in the form of hardware or in the form of a software functional module. If the above-mentioned integrated module is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. The storage medium can be a read-only memory, a disk or an optical disk, etc.

[0197] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for predicting air conditioning energy consumption, characterized in that: include: Obtain the operating data of the air conditioner, including return air temperature / humidity, ambient temperature / humidity and collection time; The operating data is subjected to sequence processing to determine a target sequence, including: converting the operating data into a regular sequence according to a specified time interval, calculating and determining the autocorrelation coefficient of the regular sequence according to different time steps, performing hypothesis testing according to the autocorrelation coefficients, and determining the target sequence, specifically: determining the corresponding p value from a probability density function table of a t test according to each of the autocorrelation coefficients, determining the maximum autocorrelation coefficient corresponding to the p value less than a specified significance level, taking the time step corresponding to the maximum autocorrelation coefficient as the target time step, and determining the target sequence according to the target time step and the regular sequence; Performing a moving weighted calculation on the target sequence through a preset algorithm to determine a weighted moving mean includes: determining a moving weight corresponding to each data value in the target sequence, performing a weighted calculation based on the moving weight and the data value to obtain a weighted moving mean, the formula being: in, is time t i The weighted moving mean of , i is the sequence number of the time point of the regular sequence, j is the sequence number of the time point of the running data, w is the moving weight, is the time t in the regular sequence i-j The moving weights corresponding to the data values ​​gradually increase with the time sequence of the data values; The target energy consumption is predicted according to the weighted moving average, the energy consumption prediction model and the recursive strategy.

2. The air conditioning energy consumption prediction method according to claim 1, characterized in that: The predicting target energy consumption according to the weighted moving mean, the energy consumption prediction model and the recursive strategy includes: Determining predicted energy consumption according to the weighted moving mean and the energy consumption prediction model; Returning to the step of obtaining the operating data of the air conditioner to determine a new predicted energy consumption, until all predicted energy consumptions for the specified observation time period are obtained; Sum up all predicted energy consumptions to get the target energy consumption.

3. The air conditioning energy consumption prediction method according to claim 2, characterized in that: Determining the predicted energy consumption according to the weighted moving mean and the energy consumption prediction model includes: Inputting the weighted moving mean into the energy consumption prediction model, and performing inverse conversion processing on the output of the energy consumption prediction model to obtain the predicted energy consumption; or, The weighted moving average is input into the energy consumption prediction model, and the output value of the energy consumption prediction model is used as the predicted energy consumption.

4. The air conditioning energy consumption prediction method according to claim 1, characterized in that: The energy consumption prediction model is trained by the following steps: Get training data; The energy consumption prediction network is trained by the GBDT algorithm and the training data, and a hyperparameter combination is generated by the Latin hypercube sampling method during the training process, and the performance index of the energy consumption prediction network is obtained by participating in the training through the hyperparameter combination; The energy consumption prediction model is determined based on the hyperparameter combination corresponding to the best performance indicator.

5. An air conditioning energy consumption prediction device, characterized in that: include: The acquisition module is used to obtain the operating data of the air conditioner, including the return air temperature / humidity, ambient temperature / humidity and acquisition time; A sequence module is used to perform sequence processing on the operation data to determine a target sequence, including: converting the operation data into a regular sequence according to a specified time interval, calculating and determining the autocorrelation coefficient of the regular sequence according to different time steps, performing hypothesis testing according to the autocorrelation coefficients, and determining the target sequence, specifically: determining the corresponding p value from the probability density function table of the t test according to each of the autocorrelation coefficients, determining the maximum autocorrelation coefficient corresponding to the p value less than the specified significance level, taking the time step corresponding to the maximum autocorrelation coefficient as the target time step, and determining the target sequence according to the target time step and the regular sequence; The calculation module is used to perform a moving weighted calculation on the target sequence through a preset algorithm to determine a weighted moving mean, including: determining a moving weight corresponding to each data value in the target sequence, performing a weighted calculation based on the moving weight and the data value to obtain a weighted moving mean, the formula is: in, is time t i The weighted moving mean of , i is the sequence number of the time point of the regular sequence, j is the sequence number of the time point of the running data, w is the moving weight, is the time t in the regular sequence i-j The moving weights corresponding to the data values ​​gradually increase with the time sequence of the data values; The prediction module is used to predict the target energy consumption according to the weighted moving mean, the energy consumption prediction model and the recursive strategy.

6. An electronic device, characterized in that: include: A processor and a memory, wherein the memory stores instructions, and the instructions are loaded and executed by the processor to implement the method according to any one of claims 1 to 4.

7. A computer-readable storage medium, wherein a computer program is stored in the computer-readable storage medium, and when the computer program is executed, the method according to any one of claims 1 to 4 is implemented.

Images