Temperature regulation demand prediction method and device, electronic equipment and storage medium

Abstract

The invention relates to the technical field of computer models, and provides a temperature adjustment demand prediction method and device, electronic equipment and a storage medium. The method comprises the steps of obtaining thermal load data used for influencing a temperature adjustment requirement in a target area, and performing feature extraction on the thermal load data to obtain input thermal load features; dividing the input thermal load feature into a plurality of sub-thermal load features, and performing feature fusion on each sub-thermal load feature to obtain a dense thermal load feature; according to the method, the input thermal load feature is divided into the plurality of sub-thermal load features, and then the plurality of sub-thermal load features are fused into the dense thermal load feature, so that the generation efficiency and stability of the dense thermal load feature are ensured; and subsequently, the temperature adjustment demand of the target area is predicted based on the dense thermal load characteristics, and the accuracy and processing efficiency of the predicted temperature adjustment demand are improved.

Description

Technical Field

[0001] The present application relates to the field of computer model technology, and in particular to a temperature regulation demand prediction method and device, an electronic device, and a storage medium. Background Art

[0002] Carbon neutrality refers to the total amount of carbon dioxide or greenhouse gas emissions, directly or indirectly generated by a country, enterprise, product, activity, or individual over a given period of time. Through tree planting, energy conservation, and emission reduction, these emissions can be offset, achieving a positive balance and a relative "zero emissions" outcome. In the process of advancing industrial digitalization in heating systems, how to achieve energy conservation and emission reduction, and thus carbon neutrality, is a current research hotspot.

[0003] When centralizing heating and cooling systems, it is necessary to consider parameters such as the flow of people in the target area, indoor temperature, outdoor temperature, humidity, etc., and jointly predict the temperature regulation needs of the target area, such as increasing the cooling capacity, increasing the heating capacity, etc.

[0004] Among the related technologies, one is to make predictions through traditional machine learning. However, due to insufficient exploration of heating and cooling characteristics, the prediction accuracy is low, and there are technical problems such as difficulty in model construction and high complexity coefficient in solving the problem. As a result, the predicted temperature adjustment requirements are difficult to meet actual needs, resulting in a poor user experience. Summary of the Invention

[0005] In view of this, the embodiments of the present application provide a temperature adjustment demand prediction method and device, an electronic device and a storage medium to solve the problem in the prior art that the predicted temperature adjustment demand is difficult to meet the actual demand and the user experience is poor.

[0006] In a first aspect of an embodiment of the present application, a temperature adjustment demand prediction method is provided, the method comprising: obtaining heat load data in a target area that affects the temperature adjustment demand, and performing feature extraction on the heat load data to obtain an input heat load feature; dividing the input heat load feature into multiple sub-heat load features, and performing feature fusion on the sub-heat load features to obtain a dense heat load feature; and predicting the temperature adjustment demand of the target area based on the dense heat load feature.

[0007] In a second aspect of an embodiment of the present application, a temperature adjustment demand prediction device is provided, which includes: an acquisition module for acquiring heat load data in a target area that affects the temperature adjustment demand, and performing feature extraction on the heat load data to obtain an input heat load feature; a fusion module for dividing the input heat load feature into multiple sub-heat load features, and performing feature fusion on the sub-heat load features to obtain a dense heat load feature; and a prediction module for predicting the temperature adjustment demand of the target area based on the dense heat load feature.

[0008] According to a third aspect of an embodiment of the present application, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.

[0009] According to a fourth aspect of the embodiments of the present application, a computer-readable storage medium is provided, which stores a computer program. When the computer program is executed by a processor, the steps of the above method are implemented.

[0010] Compared with the prior art, the embodiments of the present application have the following beneficial effects: in the embodiments of the present application, heat load data affecting the temperature control demand in the target area is obtained, and features of the heat load data are extracted to obtain input heat load features; the input heat load features are divided into multiple sub-heat load features, and the features of each sub-heat load feature are fused to obtain a dense heat load feature; the temperature control demand of the target area is predicted based on the dense heat load feature. The present application divides the input heat load feature into multiple sub-heat load features and then fuses the multiple sub-heat load features into a dense heat load feature, thereby ensuring the generation efficiency and stability of the dense heat load feature. Subsequently, the temperature control demand of the target area is predicted based on the dense heat load feature, thereby improving the accuracy and processing efficiency of the predicted temperature control demand, and avoiding the problem in the related art that the predicted temperature control demand is difficult to meet actual needs and the user experience is poor. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the embodiments or descriptions of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0012] Figure 1 This is a flow chart of a temperature regulation demand prediction method provided in an embodiment of the application;

[0013] Figure 2 This is a flow chart of another temperature adjustment demand prediction method provided in an embodiment of the present application;

[0014] Figure 3 This is a flow chart of another temperature adjustment demand prediction method provided in an embodiment of the present application;

[0015] Figure 4 1 is a flow chart of another method for predicting temperature regulation demand provided in an embodiment of the present application;

[0016] Figure 5This is a flow chart of another temperature adjustment demand prediction method provided in an embodiment of the present application;

[0017] Figure 6 This is a flow chart of another optional temperature adjustment demand prediction method provided in an embodiment of the present application;

[0018] Figure 7 This is a basic schematic diagram of a temperature regulation demand prediction model provided in an embodiment of the application;

[0019] Figure 8 This is a schematic diagram of the structure of a temperature adjustment demand prediction device provided in an embodiment of the present application;

[0020] Figure 9 This is a structural diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0021] In the following description, specific details such as specific system structures and techniques are provided for purposes of illustration rather than limitation to facilitate a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid obscuring the description of the present application with unnecessary detail.

[0022] A method and device for predicting temperature adjustment demand according to an embodiment of the present application will be described in detail below with reference to the accompanying drawings.

[0023] Figure 1 This is a flow chart of a temperature regulation demand prediction method provided in an embodiment of the present application. Figure 1 As shown, the temperature adjustment demand forecasting method includes:

[0024] S101, obtaining heat load data in a target area that affects temperature regulation requirements, and extracting features from the heat load data to obtain input heat load features;

[0025] S102, dividing the input heat load feature into multiple sub-heat load features, and fusing the sub-heat load features to obtain a concentrated heat load feature;

[0026] S103: Predicting the temperature adjustment demand of the target area based on the intensive heat load characteristics.

[0027] It can be understood that the above-mentioned target areas are areas where there is a need for temperature regulation. For example, temperature regulation equipment (such as air conditioning, heating system, etc.) is installed in a shopping mall. In order to accurately control the air conditioning in the shopping mall and avoid overheating or overcooling, there is a need for temperature regulation in the shopping mall; for another example, temperature regulation equipment is installed in an office building. In order to accurately control the air conditioning in the office building and avoid overheating or overcooling, there is a need for temperature regulation in the office building.

[0028] It can be understood that the flow of people in the target area, indoor temperature, outdoor temperature, humidity, target temperature (the target temperature is the obtained comfortable temperature that matches the target area), etc. will all affect the temperature regulation demand of the target area. For example, when the weather is hot and the flow of people in the target area is large, the temperature regulation demand needs to increase the cooling capacity to make the temperature in the target area suitable. For another example, when the weather is hot and the flow of people in the target area is small, the temperature regulation demand needs to reduce the cooling capacity to make the temperature in the target area suitable. Based on the above reasons, it can be seen that the heat load data that affects the temperature regulation demand includes but is not limited to the flow of people in the target area, indoor temperature, outdoor temperature, humidity, and target temperature.

[0029] In some examples, to facilitate subsequent processing of the heat load data, the present application performs feature extraction on the heat load data to obtain input heat load features. Specifically, a linear transformation is performed on the heat load data, and after the linear transformation, the dimension is mapped to a specified dimension to obtain the input heat load features, thereby preparing for subsequent processing of the input heat load features.

[0030] It can be understood that in order to avoid the problem that the amount of data of the input thermal load feature is too large, directly processing the input thermal load feature will reduce the stability and efficiency of the processing. In this application, the input thermal load feature is divided into multiple sub-thermal load features. For example, this application transforms the dimension of the input thermal load feature into a specified output dimension for subsequent use, and then divides the input thermal load feature (large batch sample) of the specified output dimension into multiple sub-thermal load features (virtual small batches (ghost batch)). For example, assuming that the input tensor batchsize of the input thermal load feature of the specified output dimension is N, the size s of each sub-batch set in advance is obtained, and the number of sub-thermal load features k=N / S.

[0031] In some examples, the present application fuses the sub-heat load features to obtain a dense heat load feature, thereby ensuring the stability and processing efficiency of the dense heat load feature. Subsequently, the temperature adjustment requirements of the target area are predicted based on the dense heat load feature, thereby improving the accuracy and processing efficiency of the predicted temperature adjustment requirements.

[0032] According to the technical solution provided in the embodiments of the present application, heat load data in a target area that affects the temperature regulation demand is obtained, and features are extracted from the heat load data to obtain an input heat load feature; the input heat load feature is divided into multiple sub-heat load features, and the features of each sub-heat load feature are fused to obtain a dense heat load feature; the temperature regulation demand of the target area is predicted based on the dense heat load feature. The present application divides the input heat load feature into multiple sub-heat load features and then fuses the multiple sub-heat load features into a dense heat load feature, thereby ensuring the generation efficiency and stability of the dense heat load feature. Subsequently, the temperature regulation demand of the target area is predicted based on the dense heat load feature, thereby improving the accuracy and processing efficiency of the predicted temperature regulation demand, and avoiding the problem in related technologies that the predicted temperature regulation demand is difficult to meet actual needs and the user experience is poor.

[0033] In some embodiments, as Figure 2 As shown in the figure, each sub-heat load feature is fused to obtain a dense heat load feature, including:

[0034] S201, fusing the sub-heat load features to obtain an initial dense heat load feature;

[0035] S202, performing nonlinear mapping on the initial dense heat load features to obtain latent space features;

[0036] S203. Randomly discard neurons on the latent space features to obtain dense heat load features.

[0037] It can be understood that in order to eliminate the dimensional influence between different sub-heat load characteristics and make different sub-heat load characteristics at the same order of magnitude, different sub-heat load characteristics will be accurately fused in the future. In the step of fusing the characteristics of each sub-heat load characteristic to obtain the initial dense heat load characteristic, this application will first normalize each sub-heat load characteristic, and then perform feature fusion on the characteristics obtained after the normalization of each sub-heat load characteristic to obtain the initial dense heat load characteristic.

[0038] For example, the present application performs normalization by calculating the mean and variance of each sub-heat load feature, as follows:

[0039]

[0040] in, is the normalized data of the g-th sub-heat load characteristic, A certain characteristic value of the i-th sample in the g-th sub-heat load characteristic (before normalization), μ (g) , σ 2(g)are the mean and variance of the g-th heat load feature, and ∈ is a numerical stability constant used to avoid the denominator being zero.

[0041] In summary, for a large batch of input heat load characteristics X, its g-th sub-heat load characteristics X (g) The normalization process is:

[0042]

[0043] Where: γ is a learnable scaling parameter, β is a learnable offset parameter, and finally, all is merged into the complete output. It can be understood that the learnable parameters (such as the scaling parameter γ and the offset parameter β) are updated through back propagation.

[0044] It can be understood that the features obtained after the normalization processing of each sub-heat load feature are subjected to feature fusion to obtain the feature fusion in the initial dense heat load feature, including but not limited to feature splicing and weighted summation. Preferably, the present application performs feature splicing on the features obtained after the normalization processing of each sub-heat load feature to obtain the initial dense heat load feature, and then all the features obtained after the normalization processing are spliced ​​back to the shape of the original input heat load feature.

[0045] In some examples, the present application performs nonlinear mapping on the initial dense heat load features to obtain latent space features; illustratively, the initial dense heat load features are nonlinearly mapped through LeakyReLU to obtain latent space features, wherein LeakyReLU is an activation function, which is a variant of the ReLU activation function. LeakyReLU does not directly return the negative input to zero, but maintains the gradient flow of negative values ​​through a small negative slope.

[0046] In some examples, to improve the robustness and generalization ability of this method, this application also randomly discards neurons (Dropout) on the latent space features to obtain dense heat load features. In the initial training stage, this method still injects Gaussian noise into the latent space features after random discarding neurons to further improve the robustness and generalization ability of the model.

[0047] It can be understood that the present application can also directly use the potential space characteristics as the intensive heat load characteristics.

[0048] According to the technical solution provided in the embodiment of the present application, the sub-heat load features are feature fused to obtain an initial dense heat load feature; the initial dense heat load feature is nonlinearly mapped to obtain a latent space feature; the latent space feature is randomly discarded by neurons to obtain a dense heat load feature. This method improves the stability and efficiency of feature extraction by feature fusion of the sub-heat load features to obtain an initial dense heat load feature, and the sub-heat load features are feature fused to obtain an initial dense heat load feature, so that the initial dense heat load feature is returned to the shape of the original input heat load feature. The initial dense heat load feature is subsequently nonlinearly mapped to obtain a latent space feature; the latent space feature is randomly discarded by neurons to obtain a dense heat load feature, which improves the robustness and generalization ability of the dense heat load feature.

[0049] In some embodiments, as Figure 3 As shown, before predicting the temperature adjustment demand of the target area based on the intensive heat load characteristics, the method further includes:

[0050] S301: If the iteration number of the dense heat load is lower than a preset iteration threshold, the dense heat load feature is concatenated with the input heat load feature to obtain a first concatenated heat load feature.

[0051] S302: Obtain a new dense heat load feature based on the first spliced ​​heat load feature, until the number of iterations of the new dense heat load feature is no less than an iteration threshold.

[0052] It is understandable that in order to ensure the accuracy of the prediction results of subsequent temperature adjustment requirements, this application needs to iterate the dense heat load characteristics multiple times. It is understandable that when the dense heat load characteristics are first obtained, the number of iterations of the dense heat load characteristics is recorded as 1.

[0053] After obtaining the dense heat load feature, this application will compare the number of iterations of the dense heat load feature with a preset iteration threshold (the iteration threshold is an integer greater than or equal to 2). If the number of iterations of the dense heat load is lower than the preset iteration threshold, the dense heat load feature will be spliced ​​with the input heat load feature to obtain a first spliced ​​heat load feature, and then a new dense heat load feature will be obtained based on the first spliced ​​heat load feature.

[0054] Continuing with the above example, the steps of obtaining a new dense heat load feature based on the first spliced ​​heat load feature include: dividing the first spliced ​​heat load feature into multiple sub-heat load features, and fusing the features of each sub-heat load feature to obtain a dense heat load feature; wherein, dividing the first spliced ​​heat load feature into multiple sub-heat load features, and fusing the features of each sub-heat load feature to obtain a dense heat load feature includes fusing the features of each sub-heat load feature to obtain an initial dense heat load feature; performing nonlinear mapping on the initial dense heat load feature to obtain a latent space feature; and performing random discarding neuron processing on the latent space feature to obtain a dense heat load feature. wherein, the above-mentioned specific feature fusion, nonlinear mapping, and random discarding neuron processing are the same as the steps of "dividing the input heat load feature into multiple sub-heat load features, and fusing the features of each sub-heat load feature to obtain a dense heat load feature", and will not be repeated here.

[0055] After obtaining a new dense heat load feature, the number of iterations of the new dense heat load feature is increased by 1 until the number of iterations of the obtained dense heat load feature is not less than the iteration threshold, and then the temperature adjustment demand of the target area is predicted based on the dense heat load feature.

[0056] According to the technical solution provided in the embodiment of the present application, if the number of iterations of the dense heat load is lower than a preset iteration threshold, the dense heat load feature is spliced ​​with the input heat load feature to obtain a first spliced ​​heat load feature; a new dense heat load feature is obtained based on the first spliced ​​heat load feature until the number of iterations of the new dense heat load feature is not lower than the iteration threshold. This method ensures that historical processing results are not discarded by fully reusing the input heat load feature and the dense heat load feature obtained each time. Each round of iteration gradually corrects the noise or deviation in the dense heat load feature through feature splicing and fusion, thereby improving the accuracy of the dense heat load feature, so that the temperature adjustment demand subsequently predicted based on the dense heat load feature is more accurate.

[0057] In some embodiments, as Figure 4 As shown, the temperature adjustment demand of the target area is predicted based on the intensive heat load characteristics, including:

[0058] S401, performing feature splicing on the dense heat load feature and the input heat load feature to obtain a second spliced ​​heat load feature;

[0059] S402: Perform feature transformation on the second spliced ​​heat load feature to obtain a transformed heat load feature, and predict the temperature adjustment demand of the target area based on the transformed heat load feature.

[0060] It can be understood that in order to further improve the accuracy of the predicted temperature adjustment requirements, the present application will further perform feature splicing on the final dense heat load feature and the input heat load feature to obtain a second spliced ​​heat load feature, thereby further fully reusing the original input heat load feature.

[0061] After obtaining the second spliced ​​heat load feature, feature transformation is performed to obtain a transformed heat load feature, and the temperature adjustment demand of the target area is predicted based on the transformed heat load feature. It can be understood that the above feature transformation is a linear transformation.

[0062] According to the technical solution provided in the embodiment of the present application, the dense heat load feature and the input heat load feature are feature spliced ​​to obtain a second spliced ​​heat load feature; the second spliced ​​heat load feature is feature transformed to obtain a transformed heat load feature, and the temperature adjustment demand of the target area is predicted based on the transformed heat load feature. The above steps further reuse the original input heat load feature to ensure that the original feature is not discarded, so that the obtained second spliced ​​heat load feature is more accurate, and thus the subsequent temperature adjustment demand predicted based on the second spliced ​​heat load feature is more accurate.

[0063] In some embodiments, as Figure 5 As shown, the temperature adjustment demand of the target area is predicted based on the changing heat load characteristics, including:

[0064] S501, performing dimensional transformation on the input heat load feature to obtain a target dimensional heat load feature;

[0065] S502: Fusing the transformed heat load feature and the target dimension heat load feature to obtain a target fusion feature. The target fusion feature is used to predict the temperature adjustment demand of the target area.

[0066] It is understandable that in order to further improve the accuracy of temperature regulation requirements, this application will also perform a dimensional transformation on the input heat load feature to obtain a target dimensional heat load feature. Specifically, the input heat load feature is linearly transformed using learnable weights and biases to dynamically adjust the importance of the input feature, and activated by LeakyReLU to obtain the target dimensional heat load feature. LeakyReLU is an activation function that is a variant of the ReLU activation function. LeakyReLU does not directly return to zero for negative inputs, but instead maintains the gradient flow of negative values ​​through a small negative slope.

[0067] After obtaining the target dimension heat load feature, this application will perform feature fusion on the transformed heat load feature and the target dimension heat load feature to obtain the target fusion feature. The target fusion feature is used to predict the temperature regulation demand of the target area. Specifically, the above-mentioned feature fusion methods include but are not limited to splicing and weighted summation. Exemplarily, this application will perform weighted summation on the transformed heat load feature and the target dimension heat load feature to obtain the target fusion feature, that is, the target fusion feature = Mix×SkipOutput+(1-Mix)×MainOutput, where SkipOutput is the transformed heat load feature and MainOutput is the target dimension heat load feature. Mix is ​​the weight of the transformed heat load feature (the weight is a learnable parameter), and 1-Mix is ​​the weight of the target dimension heat load feature.

[0068] According to the technical solution provided in the embodiment of the present application, the input heat load feature is dimensional transformed to obtain the target dimensional heat load feature; the transformed heat load feature and the target dimensional heat load feature are feature fused to obtain the target fusion feature, and the target fusion feature is used to predict the temperature adjustment demand of the target area. This step jumps the target dimensional heat load feature and the transformed heat load feature, provides dynamic feature weighting, solves the gradient vanishing problem, and subsequently predicts the temperature adjustment demand based on the target fusion feature, further improving the accuracy of the predicted temperature adjustment demand.

[0069] In some embodiments, as Figure 6 As shown, before dividing the input heat load characteristic into a plurality of sub-heat load characteristics, the method further includes:

[0070] S601, performing linear transformation on the input heat load characteristics, and activating the input heat load characteristics after the linear transformation;

[0071] S602: Perform random neuron discarding processing on the activated input heat load feature to obtain a pre-processed input heat load feature.

[0072] It can be understood that in order to improve the processing efficiency of subsequent S102 to S103, the present application will preprocess the input heat load features, wherein the preprocessing includes two parts: LeakyGate and Dropout. LeakyGate is a feature transformation module, which includes linear transformation and activation process. It uses learnable weights and biases to dynamically adjust the importance of input features and is activated through LeakyReLU. LeakyReLU is an activation function, which is a variant of the ReLU activation function. LeakyReLU does not directly return to zero for negative inputs, but maintains the gradient flow of negative values ​​through a small negative slope. Dropout randomly discards the input in the first layer with a discard rate of 0.1. The present application uses the above method to achieve linear transformation of the input heat load features, activate the input heat load features after the linear transformation, and randomly discard neurons for the activated input heat load features to obtain the steps of preprocessing the input heat load features.

[0073] According to the technical solution provided in the embodiment of the present application, the input heat load characteristics are linearly transformed, and the input heat load characteristics after the linear transformation are activated; the activated input heat load characteristics are randomly discarded neurons to obtain the preprocessed input heat load characteristics, thereby improving the processing efficiency of subsequent steps.

[0074] In order to better understand the present application, the present application provides a more specific embodiment for illustration. The present application provides a temperature adjustment demand prediction model, which can implement or partially implement the above steps S101 to S103. Figure 7 As shown, Figure 7 The figure shows the basic structure of the temperature regulation demand prediction model, which includes an input layer, a preprocessing layer, two densely connected blocks (the number of densely connected blocks is flexibly set by relevant personnel, and the number of densely connected blocks is the same as the number of iterations of the dense heat load feature), an output fusion layer, and an output layer.

[0075] Among them, the input feature layer receives the heat load data, and after the initial linear transformation, maps the dimension to the specified dimension to obtain the input heat load feature, preparing for subsequent feature processing.

[0076] The temperature control demand prediction model inputs heat load features into the preprocessing layer, which consists of two parts: LeakyGate and Dropout. LeakyGate is a feature transformation module that includes linear transformation and activation processes. It uses learnable weights and biases to dynamically adjust the importance of input features and is activated through LeakyReLU. LeakyReLU is an activation function that is a variant of the ReLU activation function. LeakyReLU does not directly reset negative inputs to zero, but instead maintains the gradient flow of negative values ​​through a small negative slope. Dropout randomly discards inputs at the first layer with a dropout rate of 0.1.

[0077] After the preprocessing layer completes processing of the input heat load features, the preprocessed input heat load features are input into the connected dense connection blocks. Each dense connection block includes a fully connected layer, Ghost Batch Normalization, a LeakyReLU activation layer, and a Dropout layer. The fully connected layer is used to transform the input heat load feature dimension into the specified output dimension. Ghost Batch Normalization: The input large batch of input heat load feature samples is divided into multiple "virtual small batches" (ghost batch) sub-heat load features. Batch Normalization is performed on each small batch sub-heat load feature separately, and then the normalized results are spliced ​​together as the final output to solve the normalization deviation problem of large batch input heat load feature data. The workflow is as follows:

[0078] First, a large batch of input heat load features (batch size = N) is split into multiple smaller sub-batches (ghost batches) of heat load features. The size of each sub-batch heat load feature is S, that is, N = k·S, where k is the number of sub-batches.

[0079] Then calculate the mean and variance for each sub-batch and normalize them:

[0080]

[0081] in, is the normalized data, μ (g) , σ 2(g) are the mean and variance of the g-th sub-batch, ∈: numerical stability constant.

[0082] Finally, the normalized results of all sub-batch heat load features are concatenated back to the shape of the original large batch input heat load feature.

[0083] For a large batch input heat load characteristic X, its g-th sub-batch heat load characteristic X(g) The normalization process is:

[0084]

[0085] Among them, γ is a learnable scaling parameter (updated by back propagation), β is a learnable offset parameter (updated by back propagation), and finally, all are combined into the complete output.

[0086] This application also provides nonlinear mapping through LeakyReLU, performs nonlinear mapping on the initial dense heat load features to obtain latent space features, and randomly discards neurons on the latent space features to obtain dense heat load features.

[0087] like Figure 7 As shown, during the training phase of the temperature regulation demand prediction model, the present application also adds Gaussian noise to the dense heat load features, thereby improving the model robustness and generalization ability.

[0088] like Figure 7 As shown, after obtaining the dense heat load feature, if the output end of the dense connection block is connected to another dense connection block, the dense heat load feature output by the dense connection block will be feature-spliced ​​with the input heat load feature to obtain a first spliced ​​heat load feature; based on the first spliced ​​heat load feature, a new dense heat load feature is obtained, until the last dense connection block outputs the dense heat load feature to the output fusion layer.

[0089] It can be understood that the output fusion layer will perform feature splicing on the dense heat load feature output by the last densely connected block and the original input heat load feature to obtain a second spliced ​​heat load feature; then perform feature transformation on the second spliced ​​heat load feature to obtain a transformed heat load feature, and finally perform weighted fusion on the transformed heat load feature and the linearly transformed target dimension heat load feature to obtain the target fusion feature.

[0090] Specific formula:

[0091] Output=Mix×Skip Output+(1-Mix)×Main Output;

[0092] Where Output is the target fusion feature, SkipOutput is the transformed heat load feature, and MainOutput is the target dimension heat load feature. Mix is ​​the weight of the transformed heat load feature (this weight is a learnable parameter), and 1-Mix is ​​the weight of the target dimension heat load feature.

[0093] Finally, the temperature adjustment demand prediction model outputs the temperature prediction results based on the above target fusion features.

[0094] This application proposes an efficient and lightweight temperature control demand prediction model, and the neural network architecture of the temperature control demand prediction model has made innovative progress in feature reuse, gradient propagation and noise robustness. The Ghost Batch Normalization in the neural network architecture can provide virtual batch normalization to optimize the efficiency and stability of large-batch training. The dense connection dynamically splices the input in each block to ensure full feature reuse. The output fusion layer is combined with LeakyGate to provide a dynamic feature weighting mechanism to solve the gradient disappearance problem. Gaussian noise and normalization work together to improve the noise robustness and generalization ability of the neural network architecture.

[0095] It is understandable that the above neural network architecture can also be used in scenarios such as financial prediction, customer churn prediction, and click-through rate prediction, which will not be elaborated here.

[0096] In some embodiments, after predicting the temperature control demand of the target area based on the intensive heat load characteristics, the method further includes: controlling the temperature control equipment based on the temperature control demand, thereby achieving accurate control of the temperature control equipment; illustratively, when the temperature control demand is to increase the cooling capacity, the cooling capacity of the temperature control equipment needs to be increased; when the temperature control demand is to increase the heating capacity, the heating capacity of the temperature control equipment needs to be increased.

[0097] All of the above optional technical solutions can be combined in any way to form optional embodiments of the present application, and will not be described in detail here.

[0098] The following are device embodiments of the present application, which can be used to implement the method embodiments of the present application. For details not disclosed in the device embodiments of the present application, please refer to the method embodiments of the present application.

[0099] This embodiment also provides a temperature adjustment demand prediction device, such as Figure 8 As shown, the device includes:

[0100] An acquisition module 801 is used to acquire heat load data in a target area that affects temperature regulation requirements, and perform feature extraction on the heat load data to obtain input heat load features;

[0101] A fusion module 802 is used to divide the input heat load feature into multiple sub-heat load features and perform feature fusion on each sub-heat load feature to obtain a concentrated heat load feature;

[0102] The prediction module 803 is used to predict the temperature adjustment demand of the target area based on the intensive heat load characteristics.

[0103] In some examples, the fusion module 802 is further used to perform feature fusion on each sub-heat load feature to obtain an initial dense heat load feature; perform nonlinear mapping on the initial dense heat load feature to obtain a latent space feature; and perform random neuron discarding processing on the latent space feature to obtain a dense heat load feature.

[0104] In some examples, the fusion module 802 is further configured to, if the number of iterations of the dense heat load feature is lower than a preset iteration threshold, perform feature splicing on the dense heat load feature and the input heat load feature to obtain a first spliced ​​heat load feature; and obtain a new dense heat load feature based on the first spliced ​​heat load feature until the number of iterations of the new dense heat load feature is no lower than the iteration threshold.

[0105] In some examples, the prediction module 803 is further used to perform feature splicing on the dense heat load feature and the input heat load feature to obtain a second spliced ​​heat load feature; perform feature transformation on the second spliced ​​heat load feature to obtain a transformed heat load feature, and predict the temperature adjustment demand of the target area based on the transformed heat load feature.

[0106] In some examples, the prediction module 803 is also used to perform dimensional transformation on the input heat load feature to obtain a target dimensional heat load feature; perform feature fusion on the transformed heat load feature and the target dimensional heat load feature to obtain a target fusion feature, and the target fusion feature is used to predict the temperature regulation demand of the target area.

[0107] In some examples, the fusion module 802 is further used to perform linear transformation on the input heat load feature, activate the linearly transformed input heat load feature, and perform random neuron discarding processing on the activated input heat load feature to obtain the preprocessed input heat load feature.

[0108] In some examples, the prediction module 803 is further configured to control a temperature regulation device based on the temperature regulation demand.

[0109] According to the technical solution provided in the embodiments of the present application, heat load data in a target area that affects the temperature regulation demand is obtained, and features are extracted from the heat load data to obtain an input heat load feature; the input heat load feature is divided into multiple sub-heat load features, and the features of each sub-heat load feature are fused to obtain a dense heat load feature; the temperature regulation demand of the target area is predicted based on the dense heat load feature. The present application divides the input heat load feature into multiple sub-heat load features and then fuses the multiple sub-heat load features into a dense heat load feature, thereby ensuring the generation efficiency and stability of the dense heat load feature. Subsequently, the temperature regulation demand of the target area is predicted based on the dense heat load feature, thereby improving the accuracy and processing efficiency of the predicted temperature regulation demand, and avoiding the problem in related technologies that the predicted temperature regulation demand is difficult to meet actual needs and the user experience is poor.

[0110] Figure 9 Schematic diagram of the electronic device 9 provided in the embodiment of the present application. Figure 9 As shown, the electronic device 9 of this embodiment includes: a processor 901, a memory 902, and a computer program 903 stored in the memory 902 and executable by the processor 901. When the processor 901 executes the computer program 903, the steps of the above-described method embodiments are implemented. Alternatively, when the processor 901 executes the computer program 903, the functions of the modules / units in the above-described device embodiments are implemented.

[0111] The electronic device 9 may be a desktop computer, a notebook, a PDA, a cloud server or other electronic device. The electronic device 9 may include but is not limited to a processor 901 and a memory 902. Those skilled in the art will understand that Figure 9 This is merely an example of the electronic device 9 and does not limit the electronic device 9 . The electronic device 9 may include more or fewer components than shown in the figure, or different components.

[0112] The processor 901 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.

[0113] The memory 902 can be an internal storage unit of the electronic device 9, such as a hard disk or memory of the electronic device 9. The memory 902 can also be an external storage device of the electronic device 9, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash memory card, etc. equipped on the electronic device 9. The memory 902 can also include both an internal storage unit of the electronic device 9 and an external storage device. The memory 902 is used to store computer programs and other programs and data required by the electronic device.

[0114] Those skilled in the art will clearly understand that for the sake of convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

[0115] If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the process in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and when the computer program is executed by the processor, it can implement the steps of the above-mentioned various method embodiments. The computer program may include computer program code, which may be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0116] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present application, and should all be included in the scope of protection of the present application.

Claims

1. A temperature regulation demand forecasting method, characterized in that: The method comprises: Acquiring heat load data affecting temperature regulation requirements in a target area, and performing feature extraction on the heat load data to obtain input heat load features; Dividing the input heat load feature into multiple sub-heat load features, and fusing the sub-heat load features to obtain a dense heat load feature; The temperature adjustment demand of the target area is predicted based on the intensive heat load characteristics.

2. The method according to claim 1, characterized in that The sub-heat load features are fused to obtain a concentrated heat load feature, including: Performing feature fusion on each of the sub-heat load features to obtain an initial dense heat load feature; Performing nonlinear mapping on the initial dense heat load feature to obtain a latent space feature; The latent space feature is subjected to random neuron discarding processing to obtain the dense heat load feature.

3. The method according to claim 1, characterized in that Before predicting the temperature adjustment demand of the target area based on the intensive heat load characteristics, the method further includes: If the iteration number of the dense heat load feature is lower than a preset iteration threshold, the dense heat load feature is spliced ​​with the input heat load feature to obtain a first spliced ​​heat load feature; A new concentrated heat load signature is obtained based on the first spliced ​​heat load signature, until the number of iterations of the new concentrated heat load signature is not less than the iteration threshold.

4. The method according to claim 1, wherein Predicting the temperature adjustment demand of the target area based on the intensive heat load characteristics includes: Performing feature splicing on the intensive heat load feature and the input heat load feature to obtain a second spliced ​​heat load feature; A feature transformation is performed on the second spliced ​​heat load feature to obtain a transformed heat load feature, and a temperature adjustment demand of the target area is predicted based on the transformed heat load feature.

5. The method according to claim 4, characterized in that Predicting the temperature adjustment demand of the target area according to the transformed heat load characteristics includes: Perform dimension transformation on the input heat load feature to obtain the target dimension heat load feature; The transformed heat load feature and the target dimension heat load feature are subjected to feature fusion to obtain a target fusion feature, and the target fusion feature is used to predict the temperature adjustment demand of the target area.

6. The method according to claim 3, characterized in that Before dividing the input heat load signature into a plurality of sub-heat load signatures, the method further includes: Performing linear transformation on the input heat load characteristics, and activating the input heat load characteristics after the linear transformation; The activated input heat load feature is subjected to random neuron discarding processing to obtain the preprocessed input heat load feature.

7. The method according to claim 3, characterized in that After predicting the temperature adjustment demand of the target area based on the intensive heat load characteristics, the method further includes: controlling the temperature adjustment device based on the temperature adjustment demand.

8. A temperature adjustment demand prediction device, characterized in that: The device comprises: an acquisition module, configured to acquire heat load data in a target area that affects temperature regulation requirements, and perform feature extraction on the heat load data to obtain input heat load features; A fusion module is used to divide the input heat load feature into multiple sub-heat load features, and perform feature fusion on each of the sub-heat load features to obtain a dense heat load feature; A prediction module is used to predict the temperature adjustment demand of the target area based on the intensive heat load characteristics.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images