Dynamic adaptive optimization scheduling method and system based on hybrid deep learning, medium and equipment

Abstract

The invention relates to the field of urban water supply pipe networks, and discloses a dynamic self-adaptive optimization scheduling method and system based on hybrid deep learning, a medium and equipment, and the method comprises the steps: transmitting collected data related to water plant operation and maintenance, a pipe network hydraulic model and user water consumption to a user water consumption prediction model, updating a dynamic water consumption prediction matrix, and obtaining a user water consumption prediction model; calculating the predicted water use condition; taking a dynamic water volume matrix obtained by water volume prediction as a dynamic water volume capable of being updated, and dynamically updating the water volume, the water supply volume of a water plant and the supply pressure in the pipe network by constructing a relation between a water meter and a hydraulic model node to obtain the current pipe network operation condition so as to support daily operation and maintenance; the method comprises the following steps: predicting the water quantity, determining the total water supply quantity of a water plant, determining the number of optimization hours as n hours, taking the water supply quantity predicted by the water quantity as the water demand quantity of optimization scheduling, optimizing a water plant operation and maintenance scheme, evaluating a hydraulic state after the scheme is generated, and outputting a dynamic optimization scheduling scheme.

Description

Technical Field

[0001] This invention relates to the field of urban water supply network technology, and in particular to a dynamic adaptive optimization scheduling method, system, medium and equipment based on hybrid deep learning. Background Technology

[0002] Water supply networks are crucial urban infrastructure, and their stable and efficient operation is a vital cornerstone for ensuring people's livelihoods. Among the parameters affecting the operation of these networks, the hydraulic and water quality characteristics are paramount. Substandard hydraulic and water quality at any point can lead to user dissatisfaction or even water pollution incidents. Due to limitations in the number of monitoring points, understanding the overall hydraulic condition of the water supply network requires simulation using a hydraulic model. This involves dynamically allocating water volume to each user point, while also considering the status of water treatment plants, valves, and other infrastructure to conduct a comprehensive hydraulic and water quality simulation of the network.

[0003] Currently, due to limitations in data acquisition and model building, most hydraulic models of water supply networks are static models, meaning they are only simulated using data from certain time periods. While these models can effectively represent the hydraulic and water quality conditions of the entire urban water supply network, their static nature limits simulations to specific operating conditions. They are often only suitable for simulations such as pipeline planning, scheme analysis, and accident simulation, making them difficult to apply to routine optimization and scheduling simulations. Summary of the Invention

[0004] To address the aforementioned problems, the purpose of this invention is to provide a dynamic adaptive optimization scheduling method, system, medium, and device based on hybrid deep learning, which improves the real-time, dynamic, and reliable characteristics of water supply network optimization scheduling.

[0005] To achieve the above objectives, in a first aspect, the technical solution adopted by the present invention is as follows: a dynamic adaptive optimization scheduling method based on hybrid deep learning, comprising: taking collected data on water plant operation and maintenance, pipeline hydraulic models, and user water usage as input data, transmitting them to a user water consumption prediction model, updating a dynamic water consumption prediction matrix, obtaining water consumption prediction results, and calculating the expected water usage; and constructing a dynamic water consumption matrix by combining the water consumption prediction results with monitoring data. As a real-time water volume of the pipeline network that can be updated, the connection between water meters and hydraulic model nodes is established to dynamically update the water volume, water plant supply volume, and supply pressure in the pipeline network, thereby obtaining the current pipeline network operation status to support daily operation and maintenance; through water volume prediction, the total water supply of the water plant is determined, the optimization hours are determined to be n hours, and the predicted water supply volume is set as... The water demand is used for optimized scheduling, and the water plant operation and maintenance plan is optimized. After the plan is generated, the hydraulic state is evaluated in combination with the obtained pipeline operation status, and a dynamic optimized scheduling plan is output.

[0006] Furthermore, the user water consumption prediction model is a hybrid neural network for water consumption prediction, which combines a temporal convolutional neural network (TCN) and a long short-term memory network (LSTM).

[0007] Furthermore, TCN employs random convolutions and processes temporal data through set filters. The temporal data at time t in the upper layer only comes from data from nearby past times. The temporal convolutional network introduces dilated convolutions and residual modules. The residual module consists of two convolutional units and one nonlinear mapping unit. Each convolutional unit includes a one-dimensional dilated causal convolution, weight normalization, a modified linear unit activation function, and a random deactivation operation, and adds a 1×1 convolution on the shortcut branch.

[0008] Furthermore, the Long Short-Term Memory (LSTM) network consists of multiple memory units, each with two states: a hidden state and a cellular state; each memory unit consists of three internal parts, enabling the LSTM network to forget, input, and output information.

[0009] Furthermore, the dynamic water volume prediction matrix is ​​as follows:

[0010] In the formula, Represents the m-th user At the nth moment in the future Predicted water volume; Furthermore, in the water consumption forecasting process, uncertainty prediction is performed on the user water consumption forecasting model. The specific process is as follows: By assuming that the water volume follows a Gaussian distribution, and simultaneously predicting the mean and standard deviation, the uncertainty of water volume prediction is characterized; when the water volume change value exceeds a set threshold, this uncertainty supports dispatchers in making a comprehensive judgment.

[0011] Furthermore, the optimized scheduling includes the following objective function and constraints: The objective function is: ; Water volume constraints are: ; The constraints for changes in pump operating status are:

[0012] In the formula, Record the power consumption of the optimized plant's operation; This represents the search space for optimal scheduling of water plants; To meet the target water supply needs of the water plant; The water supply volume for optimizing the water plant's plan; Indicates that the water pump assembly is The total water supply is Optimal energy consumption at that time; k For water pump combination One water pump in the middle; For water pump combination Remove water pump k Water pump assembly; For water pumps k Standalone operating frequency is f Water supply at that time; For water pumps k Standalone operating frequency is f Energy consumption per hour.

[0013] Secondly, the technical solution adopted by this invention is as follows: a dynamic adaptive optimization scheduling system based on hybrid deep learning, comprising: a data collection module, which takes collected data on water plant operation and maintenance, pipeline hydraulic model, and user water usage as input data, transmits it to the user water consumption prediction model, updates the dynamic water consumption prediction matrix, obtains water consumption prediction results, and calculates the expected water usage; and a dynamic hydraulic simulation module, which constructs a dynamic water consumption matrix by combining the water consumption prediction results with monitoring data. As a real-time water volume of the pipeline network that can be updated, the connection between water meters and hydraulic model nodes is established to dynamically update the water volume, water plant supply volume, and supply pressure in the pipeline network, thereby obtaining the current pipeline network operation status to support daily operation and maintenance; the optimization scheduling module, after water volume prediction, determines the total water supply of the water plant, determines the number of hours for optimization as n hours, and sets the predicted water supply volume as... The water demand is used for optimized scheduling, and the water plant operation and maintenance plan is optimized. After the plan is generated, the hydraulic state is evaluated in combination with the obtained pipeline operation status, and a dynamic optimized scheduling plan is output.

[0014] Thirdly, the technical solution adopted by the present invention is: a computer-readable storage medium for storing one or more programs, wherein the one or more programs include instructions, which, when executed by a computing device, cause the computing device to perform any of the methods described above.

[0015] Fourthly, the technical solution adopted by the present invention is: a computing device comprising: one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include instructions for performing any of the methods described above.

[0016] The present invention has the following advantages due to the adoption of the above technical solutions: This invention employs deep learning to effectively update the hydraulic model of the pipeline network, enabling it to adapt to daily scheduling needs. Simultaneously, by combining multi-time-step prediction, it forecasts the water volume in the pipeline network in advance, providing lead time and theoretical basis for optimized scheduling, supporting dynamic adaptive optimization. This provides dispatchers with reliable scheduling references and offers theoretical support for the efficiency and reliability of urban water supply networks. Attached Figure Description

[0017] Figure 1 This is a flowchart of the dynamic adaptive optimization scheduling method based on deep learning in an embodiment of the present invention; Figure 2 This is the structure of the hybrid deep learning water quantity prediction model in the embodiments of the present invention; Figure 3 This is a water volume prediction framework diagram in an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating the accuracy of user point water volume prediction in an embodiment of the present invention; Figure 5 This is a diagram showing the distribution of water consumption prediction errors for the entire pipe network in this embodiment of the invention. Figure 6 This is a simulation diagram of node pressure in an embodiment of the present invention; Figure 7 This is a simulation diagram of pipe flow rate in an embodiment of the present invention; Figure 8 This is a diagram illustrating the optimized operation of the water pump in an embodiment of the present invention. Detailed Implementation

[0018] To address the time delay in water meter transmission and the static problems of hydraulic models, and to improve the real-time, dynamic, and reliable characteristics of water supply network optimization scheduling, this invention provides a dynamic adaptive optimization scheduling method, system, medium, and equipment based on hybrid deep learning. This provides reliable scheduling reference for dispatchers and offers theoretical support for the efficiency and reliability of urban water supply networks.

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention are within the scope of protection of the present invention.

[0020] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0021] In one embodiment of the present invention, a dynamic adaptive optimization scheduling method based on hybrid deep learning is provided. In this embodiment, as... Figure 1 As shown, the method includes the following steps: 1) The collected data on water plant operation and maintenance, pipeline hydraulic model and user water use are used as input data and transmitted to the user water consumption prediction model to update the dynamic water consumption prediction matrix and obtain the water consumption prediction results in order to calculate the expected water consumption. 2) Construct a dynamic water volume matrix by combining the water volume prediction results obtained in step 1) with the monitoring data. As a real-time water volume of the pipeline network that can be updated, by establishing the connection between water meters and hydraulic model nodes, the water volume, water supply volume of water plants, and supply pressure in the pipeline network are dynamically updated to obtain the current operation status of the pipeline network and support daily operation and maintenance. 3) Based on water volume forecasting, determine the total water supply of the water plant, set the optimization timeframe to n hours, and set the forecast result (i.e., the predicted required water supply) as... The water demand is used for optimized scheduling, and the water plant operation and maintenance plan is optimized. After the plan is generated, the hydraulic state is evaluated in combination with the pipeline operation status obtained in step 2), and a dynamic optimized scheduling plan is output.

[0022] In step 1) above, the data collected regarding water plant operation and maintenance, pipeline hydraulic models, and user water usage mainly includes the following important data: (1) Data on water supply flow, pressure (pump head), and energy consumption of the water plant, and the start-up and shut-down status of the water plant pumps under the corresponding conditions.

[0023] (2) Pipeline hydraulic model, including information on nodes, pipelines, valves and other basic structures.

[0024] (3) Hourly water usage information for urban residents and non-residents.

[0025] The data on water plant supply flow, pressure, energy consumption, and pump status are stored to optimize scheduling plans and support the generation of subsequent optimized plans. The pipeline hydraulic model is used to simulate urban water supply conditions, obtaining pressure distribution and flow supply information across the entire pipeline network. Hourly water consumption information for both urban residents and non-residents is incorporated into the pipeline hydraulic model, enabling dynamic updates and, combined with a water volume prediction model, forecasting the water volume of the entire network to aid in advance scheduling decisions.

[0026] In step 1) above, the user water consumption prediction model is a hybrid neural network for water consumption prediction composed of a temporal convolutional neural network (TCN) and a long short-term memory network (LSTM).

[0027] Specifically, in water volume prediction research, Temporal Convolutional Networks (TCNs) and Long Short-Term Memory Networks (LSTMs) offer significant advantages. TCNs, based on a convolutional neural network architecture, can efficiently process large-scale water volume time-series data through temporal information extraction, expand the receptive field using dilated convolutions to capture long-term dependencies, and adaptively learn features at different time scales. LSTMs, through a unique gating mechanism, can effectively memorize long-term information, adaptively handle various changes in the sequence, and are robust to noise and missing data. Combining the two leverages TCN's ability to quickly process data and capture short-term features, while LSTM's advantage in handling long-term dependencies, effectively addressing the complexity and uncertainty of water volume changes. This significantly improves the accuracy and reliability of water volume prediction, providing strong support for real-time scheduling of urban water supply systems and long-term water resource planning and management.

[0028] In this embodiment, specifically for the model, TCN employs casual convolution, using a set filter. To process time series data Furthermore, the time series data at time t in the upper layer only comes from data from nearby past times, that is, it searches for correlations from past data. As shown in the following formula.

[0029] (1) In the formula, It is the dilated convolution at time t; This refers to the filter format and size. It is the convolution factor; yes Input data at any given time.

[0030] However, based on causal convolution, the more historical time steps of data used, the more hidden layers are involved. Consequently, the computational burden of TCN increases significantly. To overcome this limitation, temporal convolutional networks introduce dilated convolutions and residual modules to expand the search scope of historical water volume data (e.g., using fewer layers). Figure 2 As shown in the figure. Mathematically, the dilated convolution at time t is shown in the following formula.

[0031] (2) In the formula, As the expansion rate, it increases in the form of a power of 2 over the number of hidden layers. For example, if one input layer and two hidden layers are designed, then the number of hidden layers, the number of hidden layers, and the number of input layers, the number of first hidden layers, and the number of second hidden layers will all increase. The values ​​are 1, 2 and 4 respectively.

[0032] While dilated convolutions can effectively capture long-term patterns in valid historical data, this can lead to overfitting in the TCN network architecture. Therefore, residual connections are another key component of temporal convolutional networks to overcome performance degradation and avoid gradient vanishing, such as... Figure 2 As shown. The main idea of ​​residual blocks is to add the input of a temporal convolutional network block to its output, and its expression is: (3) In the formula, This represents the output after a series of transformations and activations.

[0033] Specifically, the residual module consists of two convolutional units and one nonlinear mapping unit. Each convolutional unit further includes a one-dimensional dilated causal convolution, weight normalization, a modified ReLU activation function, and a dropout operation. Additionally, a 1×1 convolution is added to the shortcut branch to ensure that the input and output of the residual module have the same dimension. This study tested the number of hidden layers and dilated convolutions, ultimately determining that 5 hidden layers are the optimal network structure, achieving the highest accuracy while maintaining fast training speed.

[0034] In this embodiment, the Long Short-Term Memory (LSTM) network consists of multiple memory units, each of which has two states: a hidden state and a cellular state. Each memory unit consists of three internal parts, enabling the LSTM network to forget, input, and output information.

[0035] Specifically, Long Short-Term Memory Network (LSTM) is a variant of Recurrent Neural Network (RNN) and is widely used in time series forecasting. The recurrent principle of RNNs is based on... Time and Add a connection between time points so that training parameters of adjacent time series data can be shared; that is, in... The weights obtained from training the hidden layer at each time step can be passed to the t layer, thereby exploring the intrinsic relationships in the time series data.

[0036] Regarding the architecture of the Long Short-Term Memory (LSTM) network, it consists of multiple memory units, such as... Figure 2 As shown, each memory unit has two states: a hidden state and a cellular state. The hidden state, which already exists in the recurrent neural network, is responsible for short-term memory. On the other hand, the cellular state, which does not exist in the recurrent neural network, has the capacity for long-term memory. Furthermore, each memory unit consists of three internal gates, which enables the Long Short-Term Memory network to forget, input, and output information. These three gates process information by determining which information needs to be removed from or retained by the cell (e.g., ...). Figure 2 As shown). The following formulas (4) to (6) explain the calculations performed.

[0037] (4) (5) (6) In the formula, and They are Input gate and forget gate for each moment; It is a candidate value; and These represent the Sigmoid activation function and the hyperbolic tangent activation function, respectively. The weight matrix represents the weight matrix of the corresponding gate; [, ] denotes the connection between two matrices; express The hidden state at any given moment; They are Input at any moment; This indicates the bias of the corresponding gate. Then, the current cell state... An update is needed; the cell state at this point determines... Based on this, forget some information and input... Some information in it.

[0038] Finally, the current output is determined by the output gate. The network calculation process is determined by formulas (7) to (9): (7) (8) (9) In the formula, yes The output gate at any given time; and They are Cellular state and hidden state at any given moment.

[0039] It can be seen that LSTM possesses a powerful ability to remember useful information and forget unnecessary information in time series prediction. In this way, Long Short-Term Memory (LSTM) networks can maximize the extraction of relationships between water volume time series data and overcome the limitation of RNNs, which can only work with... The limitation of maintaining connections between time-series hidden layers is an advantageous characteristic for long-term, highly periodic data such as water volume.

[0040] In this invention, the water volume prediction model has the following two functions, such as... Figure 3 As shown, firstly, water meter data transmission often suffers from delays of several hours, and this time lag is a key issue limiting the dynamic real-time hydraulic model. Secondly, through multi-step prediction, water volume data 24 hours in advance can be predicted, providing information on the future operational status of the pipeline for dynamic scheduling. The accuracy of water volume prediction is as follows: Figure 4 As shown, this represents the water demand prediction for one water demand point. The training and test sets perform well, and there is no overfitting. Meanwhile, the distribution of water demand prediction errors for the entire pipeline network is shown below. Figure 5 As shown, the relative error of 90% of the node predictions is less than 9%, which is a good prediction result and can meet the requirements of dynamic adaptive optimization scheduling.

[0041] Therefore, in this embodiment, a water volume prediction and dynamic update matrix is ​​designed and updated hourly. After obtaining dynamic water volume data, a hybrid prediction model is used to make predictions and update the existing dynamic water volume prediction matrix.

[0042] (10) (11) In the formula, Represents the monitored water volume matrix. Let represent the water volume of the m-th user at the k-th time in the past, where . The most recent water volume data collected for the m-th user. Represents the dynamic prediction water volume matrix. This represents the predicted water volume for the m-th user at time n in the future.

[0043] In step 1) above, there is a certain degree of uncertainty in the water volume prediction process, including not only the uncertainty of the water meter's monitoring itself, but also the prediction error, which is a significant source of uncertainty. Therefore, based on the prediction using a hybrid deep learning model, this embodiment adds the prediction of model uncertainty.

[0044] Specifically, in the water consumption forecasting process, uncertainty prediction is performed on the user water consumption forecasting model. The specific process is as follows: By assuming that the water volume follows a Gaussian distribution, and simultaneously predicting the mean and standard deviation, the uncertainty of water volume prediction is characterized; when the water volume change value exceeds a set threshold, this uncertainty supports dispatchers in making a comprehensive judgment.

[0045] (12) (13) (14) In the formula, and These are the mean and standard deviation of the water volume forecast, respectively. and They are respectively and The corresponding weights and biases It is the average value of the water meter data.

[0046] The resulting uncertainty range can effectively predict changes in user water consumption. Furthermore, within ranges of significant water consumption fluctuations, it can provide early warnings for both high and low values, offering a basis for optimized scheduling. Specifically, the uncertainties in pressure and flow simulation predictions include... Figures 6 to 7 As shown.

[0047] In step 2) above, in this embodiment, the dynamic water volume matrix obtained from water volume prediction... As a dynamic water volume that can be updated, it is worth noting that This may not represent the current water volume, as water transmission is often frequency-limited, and most long-distance water transmissions have time delays. Assuming the time difference for water transmission is z hours, then the actual current water consumption is... Generally speaking, the water supply volume and pressure of a water plant do not have a time lag, so they can be directly accessed.

[0048] Subsequently, by establishing connections between water meters and hydraulic model nodes, the water volume, water plant supply volume, and supply pressure are dynamically updated within the pipeline network, ultimately providing the current operational status of the network to support daily maintenance. The mathematical foundation of the pipeline network modeling is the conservation of energy and mass. It is assumed that the water flow in the pipeline is incompressible and conforms to the characteristics of steady-state flow. The equation for the conservation of mass in the pipeline network can be expressed as: (15) In the formula, q j Represents the node j The water flow in the connected pipes. in "Refers to the pipes that flow to the nodes," out "Refers to the pipe that flows out of the node. Furthermore, the water flow in the pipe obeys the law of conservation of energy, namely:" (16) in, H i , H j Represents a node i With nodes j water head, Indicates pipeline i - j The head loss can be calculated using the Hazen-Williams formula: (17) in, C , D and L These represent the Hazen-Williams coefficient, pipe diameter, and pipe length, respectively. The flow rate and node pressure can be solved by combining the linear equations of the energy and water conservation equations. EPANET uses an efficient solution method that adapts well to changes in pipe network structures. This embodiment uses EPANET as the tool to solve this set of equations and presents the data.

[0049] In water volume prediction, there is a confidence interval for user water volume. Therefore, this invention uses the Monte Carlo algorithm for sampling to simulate the overall flow and pressure changes and obtain the confidence intervals for flow and pressure changes. Assuming that each water volume is independent, the joint probability density function is as follows, and the probability function is given in formula (14).

[0050] (18) In step 3) above, the optimized scheduling includes the following objective function and constraints: The objective function is: (19) Water volume constraints are: (20) The constraints for changes in pump operating status are: (twenty one) In the formula, Record the power consumption of the optimized plant's operation; This represents the search space for optimal scheduling of water plants; To meet the target water supply needs of the water plant; The water supply volume for optimizing the water plant's plan; Indicates that the water pump assembly is The total water supply is Optimal energy consumption at that time; k For water pump combination One water pump in the middle; For water pump combination Remove water pump k Water pump assembly; For water pumps k Standalone operating frequency is f Water supply at that time; For water pumps k Standalone operating frequency is f The energy consumption per hour. Each time the state transition equation takes effect, it represents a change in the pump's operating state, and each pump state change is constrained to occur only 3 times within n hours.

[0051] Based on the data from the previous day, a hydraulic baseline n is determined, and the total water volume (underpressure water volume) of nodes that do not meet the water supply pressure under this baseline is calculated. (twenty two) in, This represents the total amount of water under pressure. This refers to the water volume at the under-pressure node.

[0052] Hydraulic calculations were performed on each scheme to obtain the corresponding total underpressure water volume; (twenty three) in, To optimize the total under-pressure water volume; This refers to the water volume at the under-pressure node.

[0053] Hydraulic judgment criteria: If the total water volume of nodes that do not meet the water supply pressure requirement is less than or equal to the total water volume of nodes that do not meet the target water supply pressure requirement, then the scheme is considered hydraulically feasible. (twenty four) The final optimized scheduling results of the water plant are shown in Table 1, and the optimized scheduling scheme for the water pumps is as follows: Figure 8 As shown.

[0054] In one embodiment of the present invention, a dynamic adaptive optimization scheduling system based on hybrid deep learning is provided, comprising: The data collection module takes the collected data on water plant operation and maintenance, pipeline hydraulic model and user water use as input data, transmits it to the user water consumption prediction model, updates the dynamic water consumption prediction matrix, and obtains the water consumption prediction results to calculate the expected water consumption. The dynamic hydraulic simulation module will construct a dynamic water volume matrix by combining water volume prediction results with monitoring data. As a real-time water volume of the pipeline network that can be updated, by establishing the connection between water meters and hydraulic model nodes, the water volume, water supply volume of water plants, and supply pressure in the pipeline network are dynamically updated to obtain the current operation status of the pipeline network and support daily operation and maintenance. The optimization scheduling module, after water volume prediction, determines the total water supply of the water plant, sets the optimization timeframe to n hours, and sets the predicted water supply to... The water demand is used for optimized scheduling, and the water plant operation and maintenance plan is optimized. After the plan is generated, the hydraulic state is evaluated in combination with the obtained pipeline operation status, and a dynamic optimized scheduling plan is output.

[0055] In the above embodiments, the user water consumption prediction model is a hybrid neural network for water consumption prediction composed of a temporal convolutional neural network (TCN) and a long short-term memory network (LSTM).

[0056] In this embodiment, TCN uses random convolution and processes temporal data through the set filters. The temporal data at time t in the upper layer only comes from the data of the nearest past time. The temporal convolutional network introduces dilated convolution and residual modules. The residual module consists of two convolutional units and one nonlinear mapping unit. Each convolutional unit includes a one-dimensional dilated causal convolution, weight normalization, a modified linear unit activation function, and a random deactivation operation, and adds a 1×1 convolution on the shortcut branch.

[0057] In this embodiment, the Long Short-Term Memory (LSTM) network consists of multiple memory units, each of which has two states: a hidden state and a cellular state. Each memory unit consists of three internal parts, enabling the LSTM network to forget, input, and output information.

[0058] In the above embodiments, the dynamic water volume prediction matrix is:

[0059] In the formula, Represents the m-th user At the nth moment in the future Predicted water volume; In the above embodiments, during the water consumption prediction process, uncertainty prediction is performed on the user water consumption prediction model. The specific process is as follows: By assuming that the water volume follows a Gaussian distribution, and simultaneously predicting the mean and standard deviation, the uncertainty of water volume prediction is characterized; when the water volume change value exceeds a set threshold, this uncertainty supports dispatchers in making a comprehensive judgment.

[0060] In the above embodiments, the optimized scheduling includes the following objective function and constraints: The objective function is: ; Water volume constraints are: ; The constraints for changes in pump operating status are:

[0061] In the formula, Record the power consumption of the optimized plant's operation; This represents the search space for optimal scheduling of water plants; To meet the target water supply needs of the water plant; The water supply volume for optimizing the water plant's plan; Indicates that the water pump assembly is The total water supply is Optimal energy consumption at that time; k For water pump combination One water pump in the middle; For water pump combination Remove water pump k Water pump assembly; For water pumps k Standalone operating frequency is f Water supply at that time; For water pumps k Standalone operating frequency is f Energy consumption per hour.

[0062] The system provided in this embodiment is used to execute the above-described method embodiments. For specific processes and details, please refer to the above embodiments, which will not be repeated here.

[0063] To better understand and implement this invention, the following embodiments are provided for further detailed explanation. In this embodiment, Python 3.9 software is used as the development platform for the model. This embodiment uses a water supply network in a northern city as an example. Figure 2 The flowchart of a deep learning-based dynamic adaptive optimization scheduling method is shown, and the main steps are as follows: (1) First, obtain the water plant operation and maintenance data, pipeline hydraulic model information, and user water usage information of the water supply network system to form an optimized scheduling database. This embodiment includes one water plant with a daily water supply of 200,000 tons and a total pipeline length of 508 km. (2) Select a reference day (D) that avoids holidays (such as the Spring Festival), and use its water volume as an optimization constraint, where the reference water volume for this constraint is ( The daily energy consumption is 233,000 tons, and the baseline energy consumption is 28,892 kWh / day. ); (3) Obtain the water volume information of the day before D as water volume prediction information, and then allocate the predicted water volume to the hydraulic model to obtain the relevant hydraulic operation information. (4) Based on the historical energy consumption matrix of the above-mentioned water plant, the optimal solution is found. Let be the objective function. The constraints are the total water supply from the water plant, the amount of water under pressure, and changes in pump sets. By combining and optimizing the historical energy consumption matrix, an optimized scheduling scheme is obtained. This scheme is a pre-estimated optimized scheduling situation that can support the water plant in making production arrangements, as shown in Table 1.

[0064] Table 1 Comparison before and after optimization

[0065] This invention achieves a 4.7% optimization in energy consumption while ensuring that the daily water supply error remains within 100 tons. At the same time, it optimizes the pump allocation scheme based on historical pump set operation data, providing a basis for water plant operation and maintenance.

[0066] In one embodiment of the present invention, a computing device is provided. This computing device can be a terminal and may include a processor, a communication interface, memory, a display screen, and an input device. The processor, communication interface, and memory communicate with each other via a communication bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. When the computer programs are executed by the processor, they implement the methods described in the above embodiments. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The communication interface is used for wired or wireless communication with external terminals. Wireless communication can be achieved through Wi-Fi, a management network, NFC (Near Field Communication), or other technologies. The display screen can be a liquid crystal display (LCD) or an e-ink display. The input device can be a touch layer covering the display screen, or buttons, a trackball, or a touchpad mounted on the casing of the computing device, or an external keyboard, touchpad, or mouse. The processor can call logical instructions stored in the memory.

[0067] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and sold or used as independent products, and can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0068] In one embodiment of the present invention, a computer program product is provided, the computer program product including a computer program stored on a non-transitory computer-readable storage medium, the computer program including program instructions, and when the program instructions are executed by a computer, the computer is able to perform the methods provided in the above-described method embodiments.

[0069] In one embodiment of the present invention, a non-transitory computer-readable storage medium is provided, which stores server instructions that cause a computer to perform the methods provided in the above embodiments.

[0070] The computer-readable storage medium provided in the above embodiments has a similar implementation principle and technical effect to the above method embodiments, and will not be described again here.

[0071] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0072] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0073] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0074] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A dynamic adaptive optimization scheduling method based on hybrid deep learning, characterized in that, include: The collected data on water plant operation and maintenance, pipeline hydraulic model and user water use are used as input data and transmitted to the user water consumption prediction model to update the dynamic water consumption prediction matrix and obtain water consumption prediction results to calculate the expected water consumption. A dynamic water volume matrix will be constructed by combining water volume prediction results with monitoring data. As a real-time water volume of the pipeline network that can be updated, by establishing the connection between water meters and hydraulic model nodes, the water volume, water supply volume of water plants, and supply pressure in the pipeline network are dynamically updated to obtain the current operation status of the pipeline network and support daily operation and maintenance. Based on water volume forecasting, the total water supply of the water plant is determined, the optimization time is set to n hours, and the forecasted water supply is set as... The water demand is used for optimized scheduling, and the water plant operation and maintenance plan is optimized. After the plan is generated, the hydraulic state is evaluated in combination with the obtained pipeline operation status, and a dynamic optimized scheduling plan is output.

2. The dynamic adaptive optimization scheduling method based on hybrid deep learning as described in claim 1, characterized in that, The user water consumption prediction model is a hybrid neural network for water consumption prediction, which combines a temporal convolutional neural network (TCN) and a long short-term memory network (LSTM).

3. The dynamic adaptive optimization scheduling method based on hybrid deep learning as described in claim 2, characterized in that, TCN uses random convolutions and processes temporal data through set filters. The temporal data at time t in the upper layer only comes from data from nearby past times. The temporal convolutional network introduces dilated convolutions and residual modules. The residual module consists of two convolutional units and one nonlinear mapping unit. Each convolutional unit includes a one-dimensional dilated causal convolution, weight normalization, a modified linear unit activation function, and a random deactivation operation, and adds a 1×1 convolution on the shortcut branch.

4. The dynamic adaptive optimization scheduling method based on hybrid deep learning as described in claim 2, characterized in that, The Long Short-Term Memory (LSTM) network consists of multiple memory units, each with two states: a hidden state and a cellular state. Each memory unit is composed of three internal parts, enabling the LSTM network to forget, input, and output information.

5. The dynamic adaptive optimization scheduling method based on hybrid deep learning as described in claim 1, characterized in that, The dynamic water volume prediction matrix is ​​as follows: In the formula, Represents the m-th user At the nth moment in the future The predicted water volume.

6. The dynamic adaptive optimization scheduling method based on hybrid deep learning as described in claim 1, characterized in that, In the water consumption forecasting process, uncertainty prediction is performed on the user water consumption forecasting model. The specific process is as follows: By assuming that the water volume follows a Gaussian distribution, and simultaneously predicting the mean and standard deviation, the uncertainty of water volume prediction is characterized; when the water volume change value exceeds a set threshold, this uncertainty supports dispatchers in making a comprehensive judgment.

7. The dynamic adaptive optimization scheduling method based on hybrid deep learning as described in claim 1, characterized in that, Optimized scheduling includes the following objective function and constraints: The objective function is: ; Water volume constraints are: ; The constraints for changes in pump operating status are: In the formula, Record the power consumption of the optimized plant's operation; This represents the search space for optimal scheduling of water plants; To meet the target water supply needs of the water plant; The water supply volume for optimizing the water plant's plan; Indicates that the water pump assembly is The total water supply is Optimal energy consumption at that time; k For water pump combination One water pump in the middle; For water pump combination Remove water pump k Water pump assembly; For water pumps k Standalone operating frequency is f Water supply at that time; For water pumps k Standalone operating frequency is f Energy consumption per hour.

8. A dynamic adaptive optimization scheduling system based on hybrid deep learning, characterized in that, include: The data collection module takes the collected data on water plant operation and maintenance, pipeline hydraulic model and user water use as input data, transmits it to the user water consumption prediction model, updates the dynamic water consumption prediction matrix, and obtains the water consumption prediction results to calculate the expected water consumption. The dynamic hydraulic simulation module will construct a dynamic water volume matrix by combining water volume prediction results with monitoring data. As a real-time water volume of the pipeline network that can be updated, by establishing the connection between water meters and hydraulic model nodes, the water volume, water supply volume of water plants, and supply pressure in the pipeline network are dynamically updated to obtain the current operation status of the pipeline network and support daily operation and maintenance. The optimization scheduling module, after water volume prediction, determines the total water supply of the water plant, sets the optimization timeframe to n hours, and sets the predicted water supply to... The water demand is used for optimized scheduling, and the water plant operation and maintenance plan is optimized. After the plan is generated, the hydraulic state is evaluated in combination with the obtained pipeline operation status, and a dynamic optimized scheduling plan is output.

9. A computer-readable storage medium for storing one or more programs, characterized in that, The one or more programs include instructions that, when executed by a computing device, cause the computing device to perform any of the methods described in claims 1 to 7.

10. A computing device, characterized in that, include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs including instructions for performing any of the methods described in claims 1 to 7.

Images