Load prediction method, system and device based on energy station sensor data flow, storage medium and computer program product

Abstract

The invention belongs to the technical field of cold and hot energy load prediction, and particularly relates to a load prediction method, system and device based on energy station sensor data flow, a storage medium and a computer program product. The method comprises the steps of preprocessing energy station sensor data, extracting time sequence features as a historical cold and hot energy load data set, dividing the historical cold and hot energy load data set into a training set and a test set, constructing an LSTM neural network prediction model, optimizing an objective function of the LSTM neural network prediction model by adopting a random double-simplex programming method, and predicting the cold and hot energy load data set according to the objective function. And inputting the training set into an LSTM neural network prediction model optimized by a random dual-simplex programming method, training the LSTM neural network prediction model, and predicting the test set by using the trained LSTM neural network prediction model to obtain a predicted cold and hot energy load. The energy efficiency management level of the energy station system is improved, the energy load is accurately predicted, the energy supply is optimized, and the overall stability of the system is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of cold and hot energy load prediction, and specifically relates to a load prediction method, system, device, storage medium and computer program product based on energy station sensor data stream. Background Art

[0002] As the complexity of energy stations increases, traditional load forecasting methods based on statistics or empirical models are gradually becoming insufficient in dealing with fluctuations and variability in energy demand. Energy stations are usually equipped with a large number of sensors that can generate continuous, real-time data streams covering a variety of parameters such as ambient temperature, humidity, energy consumption, power fluctuations, etc. How to effectively extract information from these huge data streams and combine historical data with real-time changes to accurately predict future load demand has become a key issue in current energy management systems. Existing forecasting methods have difficulty handling complex multidimensional data and nonlinear relationships. Summary of the invention

[0003] In view of this, the present invention provides a load forecasting method, system, device, storage medium and computer program product based on energy station sensor data stream. The more intelligent LSTM and SDSP combined method can significantly improve the accuracy and efficiency of load forecasting.

[0004] The technical solution adopted by the present invention is: a load forecasting method based on energy station sensor data stream, comprising:

[0005] The energy station sensor data is preprocessed, and the time series features are extracted as the historical cold and hot energy load data set. The historical cold and hot energy load data set is divided into a training set and a test set. An LSTM neural network prediction model is constructed, and the objective function of the LSTM neural network prediction model is optimized by the random double simplex programming method. The training set is input into the LSTM neural network prediction model optimized by the random double simplex programming method, and the LSTM neural network prediction model is trained. The trained LSTM neural network prediction model is used to predict the test set to obtain the predicted cold and hot energy load.

[0006] More preferably, the energy station sensor data includes ambient temperature, humidity, energy consumption, and power fluctuation data collected by sensors.

[0007] Preferably, the energy station sensor data is preprocessed including:

[0008] Data cleaning: Perform preliminary inspection and cleaning of energy station sensor data to remove missing values, outliers, and noise data to ensure data integrity and validity;

[0009] Normalization: Standardize the data output by different sensors and use normalization method to eliminate dimensional differences and facilitate neural network processing;

[0010] Missing value processing: For missing data, interpolation method is used to fill in the missing values ​​to ensure the continuity of the data.

[0011] Preferably, time series feature extraction includes

[0012] Correlation analysis: Analyze the correlation between the sensor data of each energy station and the demand for cold and hot energy loads, screen out the features that have a greater impact on load forecasting, remove redundant features, and reduce data dimensions;

[0013] Principal Component Analysis (PCA): Perform principal component analysis on multidimensional sensor data to extract the main features, reduce the dimension of the data and retain most of the information;

[0014] Time series processing: In view of the temporal continuity of sensor data, sliding window technology is used to extract time series features.

[0015] Preferably, the LSTM neural network prediction model is constructed including

[0016] The long short-term memory network LSTM is selected as the framework of the load forecasting model. The model includes input layer, hidden layer and output layer.

[0017] Input layer: The input layer receives the energy station sensor data. Each sensor parameter corresponds to an input node, and the input data is the preprocessed sensor data stream;

[0018] Hidden layer: Processing time steps, including forget gate, input gate, output gate and state update, and finally calculating the hidden state;

[0019] Output layer: Generates the final cold and hot energy load prediction value based on the hidden state.

[0020] The above technical solutions can effectively capture time series characteristics and improve load forecasting accuracy.

[0021] Preferably, the objective function of the neural network prediction model optimized by the randomized dual simplex programming method includes:

[0022] Optimize the objective function: Based on the LSTM neural network prediction model, construct an objective function that includes error minimization and uncertainty avoidance, and use the random double simplex programming method to optimize the model parameters to minimize the prediction error while avoiding overfitting;

[0023] Random scenario processing: By introducing different random scenario simulations, the random double simplex programming method can help the model better cope with various external environmental changes and improve the robustness and reliability of load forecasting in practical applications;

[0024] Optimize constraints: In the process of cold and hot energy load forecasting, the stochastic double simplex programming method optimizes the constraints in the model (such as energy consumption limit, equipment capacity, etc.) to make the forecast results more in line with actual application needs.

[0025] In the above technical solution, in order to further improve the accuracy and robustness of load forecasting, the random double simplex programming method (SDSP) is introduced for optimization. In the energy station load forecasting, facing the complex nonlinear relationship and the randomness and uncertainty of the data, the random double simplex programming method (SDSP) can significantly improve the model performance by optimizing the objective function and constraints.

[0026] In the above technical solution, the stochastic double simplex programming method (SDSP) combines the advantages of the double simplex method with a randomization strategy, and avoids the problems of loops and inefficiency that may occur in the traditional double simplex method by randomly selecting variables entering the basis in each iteration.

[0027] The present invention also discloses a load forecasting system based on energy station sensor data stream, including a data acquisition module, a data preprocessing module, a feature extraction module, a data partitioning module, an SDSP optimization module, a network training module and a load forecasting module;

[0028] A data acquisition module, used to acquire energy station sensor data;

[0029] Data preprocessing module, used to preprocess the energy station sensor data;

[0030] A feature extraction module is used to extract time series features from the preprocessed energy station sensor data as a historical cold and hot energy load data set;

[0031] A data partitioning module is used to divide the historical cold and hot energy load data set into a training set and a test set;

[0032] SDSP optimization module, which is used to optimize the objective function of the constructed LSTM neural network prediction model using the stochastic dual simplex programming method;

[0033] The network training module is used to input the training set into the LSTM neural network prediction model optimized by the random double simplex programming method to train the LSTM neural network prediction model;

[0034] The load forecasting module is used to use the trained LSTM neural network prediction model to predict the test set and obtain the predicted cold and hot energy loads.

[0035] The present invention also discloses a device for a load prediction method based on energy station sensor data stream. The device for a load prediction method based on energy station sensor data stream includes: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, the steps of the load prediction method based on energy station sensor data stream are implemented.

[0036] The present invention also discloses a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the load prediction method based on the energy station sensor data stream are implemented.

[0037] The present invention also discloses a computer program product, including a computer program, which implements the steps of the load prediction method based on energy station sensor data stream when executed by a processor.

[0038] The present invention uses a long short-term memory network (LSTM) and a stochastic dual simplex programming method (SDSP) to perform load forecasting, improve the energy efficiency management level of the energy station system, accurately predict the energy load, thereby optimizing energy supply, reducing operating costs, and improving the overall stability of the system. The prediction of sensor data usually involves an estimate of future energy consumption based on historical monitoring data. As described in the energy station management process, temperature is an important parameter that affects control logic and performance. For the energy station digital twin system, comprehensive sensor data prediction can provide a real-time virtual model, which can improve the accuracy and practicality of the energy consumption prediction model by comparing and calibrating with actual operations. The application of the digital twin system can provide context-specific information and real-time feedback, resulting in higher accuracy and reliability, thereby improving the efficiency of the Internet of Things system. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0040] Figure 1 It is a flow chart of Embodiment 1 of a load forecasting method based on energy station sensor data stream of the present invention;

[0041] Figure 2 This is a functional module diagram of Embodiment 2 of a load forecasting system based on energy station sensor data streams according to the present invention;

[0042] Figure 3 Graph simulation results (predicted and actual values) for a period of time: (a) chilled water supply temperature, (b) chilled water return temperature. DETAILED DESCRIPTION

[0043] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present invention. However, it should be clear to those skilled in the art that the present invention 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 prevent unnecessary details from obstructing the description of the present invention.

[0044] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0045] It should also be understood that the term "and / or" used in the present description and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0046] As used in the present specification and the appended claims, the term "if" may be interpreted as "when" or "upon" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" may be interpreted as meaning "upon determination" or "in response to determining" or "upon detection of [described condition or event]" or "in response to detecting [described condition or event]", depending on the context.

[0047] In addition, in the description of the present specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0048] References to "one embodiment" or "some embodiments" etc. described in the present specification mean that one or more embodiments of the present invention include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0049] Embodiment 1

[0050] like Figure 1 As shown, a load forecasting method based on energy station sensor data stream includes:

[0051] The energy station sensor data is preprocessed, and the time series features are extracted as the historical cold and hot energy load data set. The historical cold and hot energy load data set is divided into a training set and a test set. An LSTM neural network prediction model is constructed, and the objective function of the LSTM neural network prediction model is optimized by the random double simplex programming method. The training set is input into the LSTM neural network prediction model optimized by the random double simplex programming method, and the LSTM neural network prediction model is trained. The trained LSTM neural network prediction model is used to predict the test set to obtain the predicted cold and hot energy load.

[0052] In one embodiment, the energy station sensor data includes ambient temperature, humidity, energy consumption, and power fluctuation data collected by sensors.

[0053] Preprocessing of energy station sensor data includes

[0054] Data cleaning: Perform preliminary inspection and cleaning of energy station sensor data to remove missing values, outliers, and noise data to ensure data integrity and validity;

[0055] Normalization: Standardize the data output by different sensors and use normalization method to eliminate dimensional differences and facilitate neural network processing;

[0056] Missing value processing: For missing data, interpolation method is used to fill in the missing values ​​to ensure the continuity of the data.

[0057] The energy station sensor data is normalized using the following formula:

[0058]

[0059] Among them, x is the energy station sensor data, x min and x max are the minimum and maximum values ​​in the data set, and x' is the normalized data.

[0060] Time series feature extraction includes

[0061] Correlation analysis: Analyze the correlation between the sensor data of each energy station and the demand for cold and hot energy loads, screen out the features that have a greater impact on load forecasting, remove redundant features, and reduce data dimensions;

[0062] Principal Component Analysis (PCA): Perform principal component analysis on multidimensional sensor data to extract the main features, reduce the dimension of the data and retain most of the information;

[0063] Time series processing: In view of the temporal continuity of sensor data, sliding window technology is used to extract time series features.

[0064] Specifically:

[0065] The Pearson correlation coefficient formula is used to measure the correlation between the sensor data of each energy station and the target load, select the features that are highly correlated with the load, and remove redundant data. The formula is:

[0066]

[0067] Among them, x i and i are two variables, r is the correlation coefficient, and They represent the sample means of x and y respectively, n is the number of samples, and i is the encoding of the data;

[0068] 2) Calculate the covariance matrix C:

[0069]

[0070] Among them, x i is the data point, μ is the mean, n is the number of samples, and T is the transpose of the matrix;

[0071] The principal components are obtained through eigenvalue decomposition, and the eigenvector corresponding to the maximum eigenvalue is selected as the main feature. If the first k principal components are selected, the data can be projected onto these principal components, thereby reducing the dimension of the data. The original data is projected onto the selected principal components to obtain the reduced-dimensional data:

[0072] X PCA =X centered ·V k ;

[0073] Among them, V kis a matrix consisting of the eigenvectors of the first k principal components;

[0074] X PCA : Data processed by principal component analysis (PCA) represents the new data matrix obtained after dimensionality reduction, which contains the main feature information;

[0075] X centered : The centralized data matrix refers to the matrix obtained by subtracting the mean of each column (each feature) of the original data matrix XXX.

[0076] Building an LSTM neural network prediction model includes

[0077] The long short-term memory network LSTM is selected as the framework of the load forecasting model. The model includes input layer, hidden layer and output layer.

[0078] Input layer: The input layer receives the energy station sensor data. Each sensor parameter corresponds to an input node, and the input data is the preprocessed sensor data stream;

[0079] Hidden layer: Processing time steps, including forget gate, input gate, output gate and state update, and finally calculating the hidden state;

[0080] Output layer: Generates the final cold and hot energy load prediction value based on the hidden state.

[0081] The above technical solutions can effectively capture time series characteristics and improve load forecasting accuracy.

[0082] Specifically:

[0083] 1) Input layer

[0084] Receive time series data collected by the sensor, input is x t ;

[0085] x t : Energy station sensor input data at time step t;

[0086] 2) Hidden Layer

[0087] The LSTM hidden layer processes the input data time-step by time-step by following the steps below:

[0088] 2.1) Forget Gate

[0089] Decide how much history to forget:

[0090] f t＝ σ(W f ·[h t-1 ,x t ]+b f );

[0091] t: time step;

[0092] f t : Forget gate output, value range [0,1], indicating the forgetting ratio;

[0093] W f : The weight matrix of the forget gate;

[0094] h t-1 : The hidden state of the previous time step;

[0095] x t : Input data of the current time step;

[0096] b f : Bias vector of forget gate;

[0097] σ: Sigmoid activation function, mapping values ​​to [0,1];

[0098] 2.2) Input Gate

[0099] Decide how much new information to include:

[0100] i t =σ(W i ·[h t-1 ,x t ]+b i );

[0101] σ: Sigmoid activation function, mapping values ​​to [0,1];

[0102] i t : Input gate output, value range [0,1], indicating the proportion of new information introduced; W i : weight matrix of input gate;

[0103] h t-1 : The hidden state of the previous time step;

[0104] x t : Input data of the current time step;

[0105] b i : bias vector of input gate;

[0106] 2.3) Candidate status

[0107] Generate candidate cell states for updating the current state:

[0108]

[0109] : Candidate unit status, value range [-1,1];

[0110] W c : Weight matrix of candidate states;

[0111] h t-1 : The hidden state of the previous time step;

[0112] x t : Input data of the current time step;

[0113] b c : bias vector of candidate state;

[0114] tanh: Hyperbolic tangent activation function, compressing the value to [-1,1];

[0115] 2.4) Unit status update

[0116] Adjust the current unit state according to the forget gate and input gate:

[0117]

[0118] C t : The unit state at the current time step;

[0119] C t-1 : The unit state at the previous time step;

[0120] ⊙: Element-wise multiplication (Hadamard Product);

[0121] 2.5) Output Gate

[0122] Determine which cell states participate in the generation of hidden states:

[0123] o t =σ(W o ·[h t-1 ,x t ]+b o );

[0124] o t : Output gate output, value range [0,1];

[0125] W o : The weight matrix of the output gate;

[0126] h t-1 : The hidden state of the previous time step;

[0127] x t : Input data of the current time step;

[0128] b o : bias vector of output gate;

[0129] 2.6.) Hidden State

[0130] Compute the hidden state at the current time step:

[0131] h t =o t ⊙tanh(C t );

[0132] o t : Output gate output, value range [0,1];

[0133] h t : The hidden state of the current time step, used for the calculation of the next time step;

[0134] C t : The unit state at the current time step;

[0135] tanh: Hyperbolic tangent activation function, compressing the value to [-1,1];

[0136] 3) Output layer

[0137] Through the hidden state h t , using the fully connected layer to generate load forecast values:

[0138] yt=W y ·h t +b y ;

[0139] y t : load forecast value at time step t;

[0140] W y : The weight matrix of the fully connected layer;

[0141] h t : The hidden state of the current time step, used for the calculation of the next time step;

[0142] b y : Bias vector of the fully connected layer.

[0143] The objective function of the neural network prediction model optimized by stochastic dual simplex programming method includes

[0144] Optimize the objective function: Based on the LSTM neural network prediction model, construct an objective function that includes error minimization and uncertainty avoidance, and use the random double simplex programming method to optimize the model parameters to minimize the prediction error while avoiding overfitting;

[0145] Random scenario processing: By introducing different random scenario simulations, the random double simplex programming method can help the model better cope with various external environmental changes and improve the robustness and reliability of load forecasting in practical applications;

[0146] Optimize constraints: In the process of cold and hot energy load forecasting, the stochastic double simplex programming method optimizes the constraints in the model (such as energy consumption limit, equipment capacity, etc.) to make the forecast results more in line with actual application needs.

[0147] Specifically:

[0148] Optimization objective function construction

[0149] The core of the stochastic dual simplex programming method is to optimize the objective function of the LSTM neural network prediction model. The objective function consists of prediction error and constraints, which are defined as follows:

[0150]

[0151] F(θ): objective function, which represents the function to be optimized and depends on the parameter θ;

[0152] E: refers to the expected value of a random variable;

[0153] : Prediction value With the actual value y t The error function between

[0154] λR(θ): Regularization term, used to prevent model overfitting, λ is the regularization weight;

[0155] R(θ): Regularization term, used to impose constraints on the model parameters θ, usually to prevent overfitting or improve the robustness of the model; common regularization methods include L1 regularization and L2 regularization;

[0156] θ: LSTM model parameters (weights and biases);

[0157] Random scene generation

[0158] In load forecasting optimization, in order to deal with uncertainty, the random double simplex programming method introduces multiple random scenarios for simulation; by sampling the uncertainty distribution of sensor data (such as normal distribution or uniform distribution), the forecast data stream under different scenarios is generated; the objective function is optimized in each scenario:

[0159]

[0160] s: the number of the random scene;

[0161] N: number of scene samples;

[0162] λR(θ): Regularization term, used to prevent model overfitting, λ is the regularization weight;

[0163] R(θ): Regularization term, used to impose constraints on model parameters θ to prevent overfitting or improve the robustness of the model;

[0164] θ: LSTM model parameters (weights and biases);

[0165] Dual Simplex Optimization Procedure

[0166] The SDSP method optimizes model parameters through the following steps:

[0167] Initialization: Set the initial model parameters θ 0 , randomly generate an initial scene set S 0 ;

[0168] Constraint processing: Introduce the physical limitations of the energy station (such as maximum load capacity, power fluctuation threshold, etc.) as optimization constraints:

[0169] Constraints:

[0170] G(θ)≤0;

[0171] Objective function iteration: In each scenario k, the dual simplex method is used to optimize the objective function and update the model parameters:

[0172]

[0173] θ k : The parameter vector of the current iteration step k; represents the model parameters at the kth step;

[0174] θ k+1 : The parameter vector for the next iteration, representing the updated parameters at the k+1th step;

[0175] α: learning rate;

[0176] : The gradient of the objective function;

[0177] Random scene update: dynamically adjust the scene set S according to the optimization results k , enhance the randomness and comprehensiveness of optimization;

[0178] Convergence judgment: When the objective function value changes less than the set threshold or reaches the maximum number of iterations, stop the iteration and output the optimal parameter θ * ;

[0179] Model optimization effect evaluation

[0180] The performance of the load forecasting model after SDSP optimization is evaluated by the following indicators:

[0181] Mean Squared Error (MSE):

[0182]

[0183] MSE: Mean square error, which represents the average of the differences between all predicted values ​​and the true values;

[0184] N: sample size, indicating how many data points there are in total;

[0185] y t : True value, the actual observed value at time point t;

[0186] : Prediction value, the prediction result of the model at time point t;

[0187] t: index, indicating the specific time step currently being calculated, ranging from 1 to n;

[0188] Mean Absolute Error (MAE):

[0189]

[0190] MAE: Mean absolute error, which represents the average of the absolute differences between all predicted values ​​and the true values;

[0191] N: sample size, indicating how many data points there are in total;

[0192] y t : True value, the actual observed value at time point t;

[0193] : Prediction value, the prediction result of the model at time point t;

[0194] t: index, indicating the specific time step currently being calculated, ranging from 1 to n;

[0195] Model robustness analysis: Statistical analysis of prediction errors is performed under different random scenarios, and the variance is calculated to measure the robustness of the model.

[0196] In order to further improve the accuracy and robustness of load forecasting, the randomized double simplex programming (SDSP) method is introduced for optimization. In energy station load forecasting, facing complex nonlinear relationships and the randomness and uncertainty of data, the randomized double simplex programming (SDSP) method can significantly improve the model performance by optimizing the objective function and constraints.

[0197] The stochastic double simplex programming method (SDSP) combines the advantages of the double simplex method with a randomization strategy. It avoids the problems of loops and inefficiency that may occur in the traditional double simplex method by randomly selecting the variables entering the basis in each iteration.

[0198] In one embodiment, the load size borne by the chiller is generally not directly available. There are two main methods for calculating the cooling load in engineering. One is to read the current load rate of the chiller on the data display panel of the chiller, and then obtain the rated cooling capacity of the equipment from the nameplate information of the chiller, and then calculate the load of the chiller; the other is to read the inlet and outlet water temperature and the cold water flow rate of the chiller in the cold water loop, and then calculate the load borne by the equipment. This embodiment uses the second method, which uses the predicted temperature and flow to calculate the load size:

[0199] Q load =PLR×Q rate

[0200] Q load =cm(t in -t out );

[0201] In the formula, Q load is the actual cooling capacity of the chiller, kW; PLR is the partial load rate of the chiller, %; Q loa d ,rate is the rated cooling capacity of the chiller, kW; c is the specific heat capacity of the heat exchange fluid in the pipeline, kj / (kg·c); m is the mass flow rate of the chiller loop, kg / s; t in is the cold water inlet temperature, °C; t out is the cold water outlet temperature, ℃.

[0202] Application of LSTM and SDSP methods in this embodiment in a WSHP energy station in a certain area:

[0203] The chilled water supply temperature and return water temperature monitoring records from March 22, 2024 to September 21, 2024 are used as the main data source. A one-week data set is extracted and used to predict the values ​​of the next week. A 2-week data set (June 3-16, 2024, 2.016×104min) is selected to demonstrate the results. Figure 3 The results shown compare the actual values ​​(blue line) with the predicted values ​​(red line). Figure 3 (a) and (b) show the predicted and actual values.

[0204] Figure 3 All the figures shown show that the fluctuation frequency of the predicted curve is roughly the same as that of the actual curve. It can be inferred from the evaluation results that the load forecasting method based on the energy station sensor data stream proposed in this embodiment effectively captures the complex correlation in the sequence data and retains the time relationship in the sequence data, which helps to improve the prediction accuracy and generalization ability.

[0205] The prediction results for the chilled water supply temperature show that the ME of the LSTM neural network prediction model of this embodiment is 0.1157, the accuracy is 99.81%, and the root mean square error is 1.9052. The LSTM neural network prediction model of this embodiment performs exceptionally well in predicting the chilled water supply temperature. The deviation between the predicted value and the actual value is very small, and most of the prediction errors fall within an acceptable range. The predicted chilled water return temperature ME is 0.6072, the accuracy is 96.39%, and the RMSE is 3.3179. Although the accuracy of the return water temperature is slightly lower than that of the supply air temperature, the overall prediction accuracy is still very high, indicating that the model also performs well in predicting the return water temperature.

[0206] Through the introduction of the SDSP method, the LSTM neural network prediction model of this embodiment can not only capture complex time series characteristics more accurately, but also effectively deal with data uncertainty, providing solid technical support for the efficient scheduling and management of energy stations.

[0207] Embodiment 2

[0208] like Figure 2 As shown, a load forecasting system based on energy station sensor data stream includes a data acquisition module, a data preprocessing module, a feature extraction module, a data partitioning module, an SDSP optimization module, a network training module and a load forecasting module;

[0209] A data acquisition module, used to acquire energy station sensor data;

[0210] Data preprocessing module, used to preprocess the energy station sensor data;

[0211] A feature extraction module is used to extract time series features from the preprocessed energy station sensor data as a historical cold and hot energy load data set;

[0212] A data partitioning module is used to divide the historical cold and hot energy load data set into a training set and a test set;

[0213] SDSP optimization module, which is used to optimize the objective function of the constructed LSTM neural network prediction model using the stochastic dual simplex programming method;

[0214] The network training module is used to input the training set into the LSTM neural network prediction model optimized by the random double simplex programming method to train the LSTM neural network prediction model;

[0215] The load forecasting module is used to use the trained LSTM neural network prediction model to predict the test set and obtain the predicted cold and hot energy loads.

[0216] Embodiment 3

[0217] A device for a load prediction method based on energy station sensor data stream, the device for a load prediction method based on energy station sensor data stream comprises: a memory, a processor and a computer program stored in the memory and executable on the processor, the computer program implementing the steps of the load prediction method based on energy station sensor data stream when executed by the processor.

[0218] Embodiment 4

[0219] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the load prediction method based on energy station sensor data stream.

[0220] Embodiment 5

[0221] A computer program product comprises a computer program, wherein when the computer program is executed by a processor, the steps of the load forecasting method based on energy station sensor data stream are implemented.

[0222] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0223] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0224] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0225] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0226] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit its protection scope. Although the present invention is described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that after reading the present invention, those skilled in the art can still make various changes, modifications or equivalent substitutions to the specific implementation methods of the invention, but these changes, modifications or equivalent substitutions are within the protection scope of the pending claims of the invention. The contents not described in detail in this specification belong to the prior art known to professional and technical personnel in the field.

Claims

1. A load forecasting method based on energy station sensor data stream, characterized in that: include The energy station sensor data is preprocessed, and the time series features are extracted as the historical cold and hot energy load data set. The historical cold and hot energy load data set is divided into a training set and a test set. An LSTM neural network prediction model is constructed, and the objective function of the LSTM neural network prediction model is optimized by the random double simplex programming method. The training set is input into the LSTM neural network prediction model optimized by the random double simplex programming method, and the LSTM neural network prediction model is trained. The trained LSTM neural network prediction model is used to predict the test set to obtain the predicted cold and hot energy load.

2. The load forecasting method based on energy station sensor data stream according to claim 1 is characterized in that: The energy station sensor data includes ambient temperature, humidity, energy consumption, and power fluctuation data collected by sensors.

3. The load forecasting method based on energy station sensor data stream according to claim 1 is characterized in that: Preprocessing of energy station sensor data includes Data cleaning: Perform preliminary inspection and cleaning of energy station sensor data to remove missing values, outliers, and noise data; Normalization: standardize the data output by different sensors; Missing value processing: For missing data, interpolation method is used to fill the missing values.

4. The load forecasting method based on energy station sensor data stream according to claim 1 is characterized in that: Time series feature extraction includes Correlation analysis: Analyze the correlation between the sensor data of each energy station and the demand for cold and hot energy loads, screen out the features that have a greater impact on load forecasting, and remove redundant features; Principal component analysis: Perform principal component analysis on multidimensional sensor data to extract main features; Time series processing: In view of the temporal continuity of sensor data, sliding window technology is used to extract time series features.

5. The load forecasting method based on energy station sensor data stream according to claim 1 is characterized in that: Building an LSTM neural network prediction model includes The long short-term memory network LSTM is selected as the framework of the load forecasting model. The model includes input layer, hidden layer and output layer. Input layer: The input layer receives the energy station sensor data. Each sensor parameter corresponds to an input node, and the input data is the preprocessed sensor data stream; Hidden layer: Processing time steps, including forget gate, input gate, output gate and state update, and finally calculating the hidden state; Output layer: Generates the final cold and hot energy load prediction value based on the hidden state.

6. The load forecasting method based on energy station sensor data stream according to claim 1 is characterized in that: The objective function of the neural network prediction model optimized by stochastic dual simplex programming method includes Optimize the objective function: Based on the LSTM neural network prediction model, construct an objective function that includes error minimization and uncertainty avoidance, and use the random double simplex programming method to optimize the model parameters to minimize the prediction error while avoiding overfitting; Random scene processing: by introducing different random scene simulations; Optimizing constraints: In the process of forecasting the cooling and heating energy loads, the stochastic double simplex programming method is used to optimize the constraints in the model.

7. A load forecasting system based on energy station sensor data stream, characterized in that: It includes data acquisition module, data preprocessing module, feature extraction module, data partitioning module, SDSP optimization module, network training module and load forecasting module; A data acquisition module, used to acquire energy station sensor data; Data preprocessing module, used to preprocess the energy station sensor data; A feature extraction module is used to extract time series features from the preprocessed energy station sensor data as a historical cold and hot energy load data set; A data partitioning module is used to divide the historical cold and hot energy load data set into a training set and a test set; SDSP optimization module, which is used to optimize the objective function of the constructed LSTM neural network prediction model using the stochastic dual simplex programming method; The network training module is used to input the training set into the LSTM neural network prediction model optimized by the random double simplex programming method to train the LSTM neural network prediction model; The load forecasting module is used to use the trained LSTM neural network prediction model to predict the test set and obtain the predicted cold and hot energy loads.

8. A device for a load forecasting method based on energy station sensor data stream, characterized in that: The device for the load forecasting method based on the energy station sensor data stream comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, the steps of the load forecasting method based on the energy station sensor data stream are implemented as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the load forecasting method based on the energy station sensor data stream as described in any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the load forecasting method based on energy station sensor data stream as described in any one of claims 1 to 6 are implemented.

Images