A Refined Monitoring Method for Water Levels in Different Areas of a Reservoir Based on Neural Networks

Abstract

A neural network-based method for refined water level monitoring in reservoirs first divides the reservoir catchment area into different regions. Then, historical data for each catchment area is collected. Next, neural network models are constructed for each type of outflow in each region and trained. Following this, a mathematical model for water level is built based on the catchment areas. Real-time data is then used to predict water levels, and the predicted data is input into the mathematical model to solve for the water surface changes and water level heights in each region at each finely granular time point. An alarm is triggered when the water level exceeds a set threshold. This method can more accurately depict water surface changes and water level heights, capturing water level differences within minute time intervals, thus achieving refined monitoring and alarm functions.

Description

Technical Field

[0001] This invention relates to the field of water conservancy and reservoir monitoring, and in particular to a method for refined monitoring of water levels in various areas of a reservoir based on neural networks. Background Technology

[0002] Reservoir water level monitoring is a crucial factor in water resource management, reflecting the flood peak regulation capacity of hydraulic projects. Its accuracy and timeliness directly affect the accuracy of flood control scheduling.

[0003] Most existing monitoring methods rely on manual monitoring or physical equipment, such as fiber optic temperature sensors and piezoresistive level gauges. These methods can determine the real-time water level of a reservoir to a certain extent, but they lack predictive capabilities. Existing technology CN108917876A uses neural networks to predict water levels by acquiring images. However, this method requires acquiring and processing a large number of waterline images, resulting in low processing efficiency. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention aims to propose a refined monitoring method for water levels in various areas of a reservoir based on neural networks, specifically employing the following technical solution:

[0005] The method for refined monitoring of water levels in various areas of a reservoir based on neural networks includes the following steps:

[0006] Step 1: Divide the reservoir catchment area into different catchment zones;

[0007] Step 2: Collect historical data of the reservoir's catchment area;

[0008] Step 3: Construct neural network models for each type of water outflow in each area of ​​the reservoir, and train these neural network models.

[0009] Step 4: Construct a mathematical model of water level based on the reservoir's catchment area;

[0010] Step 5: Real-time data collection of reservoir storage, inflow, and outflow of each type of water in each area at the current moment, input into the neural model in Step 3 to obtain the prediction data required in the mathematical model in Step 4.

[0011] Step 6: Use the predicted data from Step 5 to solve the mathematical model from Step 4. If the obtained regional water level is greater than the set alarm threshold, then issue an alarm and reduce the inflow until the water level in all regions is lower than the alarm threshold and the water storage capacity in each region of the reservoir is at its maximum.

[0012] In step 1, the reservoir catchment area is divided into a flood control area, a water storage area, a buffer zone, and a restricted area.

[0013] In step 2, the historical data of the reservoir catchment area includes the historical water storage of the reservoir, the historical water inflow of the reservoir, the historical water storage of the reservoir, the historical water outflow required for power generation in each area of ​​the reservoir, the historical water outflow required for energy storage, and the historical splash water outflow in each area of ​​the reservoir.

[0014] Collect historical data for at least one month, with a maximum collection interval of 1 hour for each historical data point.

[0015] In step 3, the water output types include water output required for power generation, water output required for energy storage, and splash water output;

[0016] The constructed neural network model consists of at least two neural units, each of which includes an input mechanism, an output mechanism, a filtering mechanism, and a storage mechanism.

[0017] The input mechanism satisfies the following relation:

[0018]

[0019] Among them, i t This represents the output of the input mechanism at this moment, where σ is the Sigmoid function, and x... t h is the input to the neural network. t-1 The output of the neural network at the previous time step is h. t-1 w is 0 xi w represents the weights of the input to the neural network. hi b represents the weights of the neural network output at the previous time step, which are the input mechanisms. i This is the penalty coefficient for the input mechanism; This indicates bit addition.

[0020] The output mechanism satisfies the following relationship:

[0021]

[0022] Among them, o t w represents the output result of the output mechanism at this moment. xo w represents the weights of the input to the output mechanism neural network. ho b represents the weights of the neural network output at the previous time step. o This is the penalty coefficient for the output mechanism.

[0023] The filtering mechanism satisfies the following relationship:

[0024]

[0025] Among them, f t w represents the output of the filtering mechanism at this moment.xf w represents the weights of the neural network input in the filtering mechanism. hf b represents the weights of the neural network output at the previous time step in the filtering mechanism. f This is the penalty coefficient for the filtering mechanism.

[0026] The storage mechanism satisfies the following relationship:

[0027]

[0028] Among them, C t For the output of the storage mechanism at this moment, C t-1 w represents the output of the storage mechanism from the previous time step. xc w represents the weights of the neural network input in the storage mechanism. hc b represents the weights of the neural network output at the previous time step in the storage mechanism. c This is the penalty coefficient for the storage mechanism; This indicates positional multiplication.

[0029] The output of the neural network at this moment satisfies the following relationship:

[0030]

[0031] When training a neural network model, other historical data of a different type than the data to be predicted are used as input to the model.

[0032] In step 4, the mathematical model for the water level in the storage area is as follows:

[0033]

[0034] f(xs t )+hf≤h max

[0035] Where, f(xs) t f(xb) represents the water level of the reservoir at time t; hf represents the water level of the flood control area; M represents the dimension of the model. The more dimensions there are, the higher the degree of subdivision of the water level of the reservoir. In this invention, M is taken as 30; f(xb) t () represents the water level of the buffer zone at time t; xs t This represents the simulated water volume in the water storage area at time t.

[0036] The simulated water volume in the storage area at time t satisfies the following relationship:

[0037]

[0038] Among them, xs t ST represents the simulated water volume in the storage area at time t; t Indicates the water storage capacity of the reservoir at time t; IFt This represents the inflow of water into the reservoir at time t, obtained through real-time data collection. This represents the simulated outflow from the water storage area at time t;

[0039]

[0040] in, This represents the predicted water output required for power generation from the water storage area at time t. This represents the predicted outflow of water required to supply energy storage to the water storage area at time t. PGE represents the predicted splashing water volume at time t in the water storage area; PGE represents the actual power generation efficiency, which is generally taken as 0.85.

[0041] The mathematical model for the water level in the buffer zone is:

[0042]

[0043] Where, hia represents the water level in the restricted area of ​​the reservoir, which is generally a fixed value; xb t This represents the amount of water in the buffer at time t.

[0044] The simulated water volume in the buffer zone at time t satisfies the following relationship:

[0045]

[0046] in, This represents the simulated outflow from the buffer zone at time t;

[0047]

[0048] in, This represents the predicted water output required for power generation at time t in the buffer zone; This represents the predicted outflow of water required to supply energy storage in the buffer zone at time t.

[0049] The beneficial effects of this invention are that, compared with the prior art, this invention:

[0050] 1. The constructed neural network model can better handle different types of water outflow in each area of ​​the reservoir at smaller, finer time periods, without the need for waterline images, thus achieving higher efficiency;

[0051] 2. The proposed mathematical model of water level in different areas of the reservoir can more accurately depict changes in water surface and water level height, and capture water level differences in small time periods, thereby achieving refined monitoring and alarm. Attached Figure Description

[0052] Figure 1 This is a flowchart illustrating the detailed process of the neural network-based method for refined monitoring of water levels in various areas of a reservoir disclosed in this invention.

[0053] Figure 2 This is a reservoir area division map for the refined monitoring method of water level in various areas of a reservoir based on neural networks disclosed in this invention. Detailed Implementation

[0054] The present application will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention, and should not be construed as limiting the scope of protection of the present application.

[0055] like Figure 1 As shown, the present invention provides a method for refined monitoring of water levels in various areas of a reservoir based on neural networks, comprising the following steps:

[0056] Step 1: Divide the reservoir catchment area into different catchment zones;

[0057] The reservoir's catchment area is divided into, for example... Figure 2 The four areas shown, from top to bottom, are the flood control area, the water storage area, the buffer zone, and the restricted area; those skilled in the art can also divide the reservoir's catchment area according to the specific function of the reservoir and the inflow and outflow conditions;

[0058] Step 2: Collect historical data of the reservoir's catchment area;

[0059] Historical data for the reservoir's catchment area includes the reservoir's historical water storage, historical inflow, historical water storage, historical outflow required for power generation in each area of ​​the reservoir, historical outflow required for energy storage, and historical splash-out volume in each area of ​​the reservoir.

[0060] Collect historical data for at least one month, with a maximum collection interval of 1 hour for each historical data point.

[0061] Step 3: Construct neural network models for each type of water outflow in each area of ​​the reservoir, and train these neural network models.

[0062] The types of water output include water output required for power generation, water output required for energy storage, and splash water output;

[0063] The constructed neural network model consists of at least two neural units, and the structure of the neural unit includes an input mechanism, an output mechanism, a filtering mechanism, and a storage mechanism.

[0064] The input mechanism satisfies the following relation:

[0065]

[0066] Among them, i t This represents the output of the input mechanism at this moment, where σ is the Sigmoid function, and x... t h is the input to the neural network.t-1 The output of the neural network at the previous time step is h. t-1 w is 0 xi w represents the weights of the input to the neural network. h x represents the weights of the neural network output at the previous time step, and b i This is the penalty coefficient for the input mechanism; Indicates bitwise addition;

[0067] The output mechanism satisfies the following relationship:

[0068]

[0069] Among them, o t w represents the output result of the output mechanism at this moment. xo w represents the weights of the input to the output mechanism neural network. ho b represents the weights of the neural network output at the previous time step. o This is the penalty coefficient for the output mechanism;

[0070] The filtering mechanism satisfies the following relationship:

[0071]

[0072] Among them, f t w represents the output of the filtering mechanism at this moment. xf w represents the weights of the neural network input in the filtering mechanism. hf b represents the weights of the neural network output at the previous time step in the filtering mechanism. f This represents the penalty coefficient for the filtering mechanism.

[0073] The storage mechanism satisfies the following relationship:

[0074]

[0075] Among them, C t For the output of the storage mechanism at this moment, C t-1 w represents the output of the storage mechanism from the previous time step. xc w represents the weights of the neural network input in the storage mechanism. hc b represents the weights of the neural network output at the previous time step in the storage mechanism. c This is the penalty coefficient for the storage mechanism; Indicates positional multiplication;

[0076] The output of the neural network at this moment satisfies the following relationship:

[0077]

[0078] When training a neural network model, other historical data of different types than the data to be predicted are used as input to the model; for example, when training a model for the amount of water required for power generation in each area of ​​a reservoir, historical data other than the historical amount of water required for power generation in each area of ​​the reservoir are used as input.

[0079] Step 4: Construct a mathematical model of water level based on the reservoir's catchment area;

[0080] Under normal circumstances, the water volume in flood control areas and restricted areas remains at a fixed value, and therefore the water level in flood control areas and restricted areas also remains at a fixed value.

[0081] The simulated water volume in the storage area at time t satisfies the following relationship:

[0082]

[0083] Among them, xs t ST represents the simulated water volume in the storage area at time t; t Indicates the water storage capacity of the reservoir at time t; IF t This represents the inflow of water into the reservoir at time t, obtained through real-time data collection. This represents the simulated outflow from the water storage area at time t;

[0084]

[0085] in, This represents the predicted water output required for power generation from the water storage area at time t. This represents the predicted water output required to supply energy storage to the water storage area at time t; This represents the predicted splashing water volume at time t in the water storage area; PGE represents the actual power generation efficiency, which is generally taken as 0.85.

[0086] The mathematical model for the water level in the reservoir is as follows:

[0087]

[0088] f(xs t )+hf≤h max

[0089] Where, f(xs) t f(xb) represents the water level of the reservoir at time t; hf represents the water level of the flood control area; M represents the dimension of the model. The more dimensions there are, the higher the degree of subdivision of the water level of the reservoir. In this invention, M is taken as 30; f(xb) t () represents the water level in the buffer zone at time t;

[0090] The mathematical model for the water level in the buffer zone is:

[0091]

[0092] Where, hia represents the water level in the restricted area of ​​the reservoir. Since water from the restricted area is typically not used, the water level there is usually a fixed value, determined based on the specific zoning. In this invention, the water level in the restricted area is 20% of the highest water level in the reservoir; xb t This represents the amount of water in the buffer zone at time t;

[0093] The simulated water volume in the buffer zone at time t satisfies the following relationship:

[0094]

[0095] in, This represents the simulated outflow from the buffer zone at time t;

[0096]

[0097] in, This represents the predicted water output required for power generation at time t in the buffer zone. This represents the predicted outflow of water required for energy storage at time t in the buffer zone;

[0098] In this invention, the minimum interval at each time point t can be 10 seconds;

[0099] Step 5: Real-time data collection of reservoir storage, inflow, and outflow of each type of water in each area at the current moment, input into the neural model in Step 3 to obtain the prediction data required in the mathematical model in Step 4.

[0100] In this embodiment, the predicted data required in the mathematical model of step 4 are the predicted water output required for power generation in the water storage area at time t, the predicted water output required for energy storage in the water storage area at time t, the predicted splash water output in the water storage area at time t, the predicted water output required for power generation in the buffer zone at time t, and the predicted water output required for energy storage in the buffer zone at time t.

[0101] Step 6: Use the predicted data from Step 5 to solve the mathematical model from Step 4. If the obtained regional water level is greater than the set alarm threshold, reduce the inflow until the water level in all regions is lower than the alarm threshold and the water storage capacity in each region of the reservoir is maximized.

[0102] The applicant of this invention has provided a detailed description of the embodiments of the invention in conjunction with the accompanying drawings. However, those skilled in the art should understand that the above embodiments are merely preferred embodiments of the invention. The detailed description is only intended to help readers better understand the spirit of the invention and is not intended to limit the scope of protection of the invention. On the contrary, any improvements or modifications made based on the inventive spirit of the invention should fall within the scope of protection of the invention.

Claims

1. A method for refined monitoring of water levels in various areas of a reservoir based on neural networks, characterized in that, The method for monitoring water levels in various areas of the reservoir includes the following steps: Step 1: Divide the reservoir catchment area into different catchment zones; Step 2: Collect historical data of the reservoir's catchment area; Historical data for the reservoir's catchment area includes the reservoir's historical water storage, historical inflow, historical outflow required for power generation in each area of ​​the reservoir, historical outflow required for energy storage, and historical splash-out in each area of ​​the reservoir. Collect historical data for at least one month, with a maximum collection interval of 1 hour for each historical data point. Step 3: Construct neural network models for the water output required for power generation, water output required for energy storage, and splash water output for each area of ​​the reservoir, and train these neural network models. The constructed neural network model consists of at least two neural units; Step 4: Construct a mathematical model of water level based on the reservoir's catchment area; Step 5: Real-time data collection of reservoir storage, inflow, and outflow of each type of water in each area at the current moment, input into the neural model in Step 3 to obtain the prediction data required in the mathematical model in Step 4. Step 6: Use the predicted data from Step 5 to solve the mathematical model from Step 4. If the obtained regional water level is greater than the set alarm threshold, then issue an alarm and reduce the inflow until the water level in all regions is lower than the alarm threshold and the water storage capacity in each region of the reservoir is at its maximum.

2. The method for refined monitoring of water levels in various areas of a reservoir based on neural networks according to claim 1, characterized in that: In step 1, the reservoir catchment area is divided into a flood control area, a water storage area, a buffer zone, and a restricted area.

3. The method for refined monitoring of water levels in various areas of a reservoir based on neural networks according to claim 2, characterized in that: In step 4, the mathematical model for the water level in the storage area is as follows: f(xs t )+hf≤h max Where, f(xs) t f(xb) represents the water level of the reservoir at time t; hf represents the water level of the flood control area; M represents the dimension of the model. The more dimensions there are, the more detailed the water level of the reservoir is. M is set to 30; t () represents the water level of the buffer zone at time t; xs t This represents the simulated water volume in the water storage area at time t.

4. The method for refined monitoring of water levels in various areas of a reservoir based on neural networks according to claim 2, characterized in that: The simulated water volume in the storage area at time t satisfies the following relationship: Among them, xs t ST represents the simulated water volume in the storage area at time t; t Indicates the water storage capacity of the reservoir at time t; IF t This represents the inflow of water into the reservoir at time t, obtained through real-time data collection. This represents the simulated outflow from the water storage area at time t; in, This represents the predicted water output required for power generation from the water storage area at time t. This represents the predicted water output required to supply energy storage to the water storage area at time t; The value represents the predicted splashing water volume at time t in the water storage area; PGE represents the actual power generation efficiency, which is taken as 0.

85.

5. The method for refined monitoring of water levels in various areas of a reservoir based on neural networks according to claim 3, characterized in that: The mathematical model for the water level in the buffer zone is: Where, hia is the water level in the restricted area of ​​the reservoir, which is a fixed value; xb t This represents the amount of water in the buffer at time t.

6. The method for refined monitoring of water levels in various areas of a reservoir based on neural networks according to claim 4, characterized in that: The simulated water volume in the buffer zone at time t satisfies the following relationship: in, This represents the simulated outflow from the buffer zone at time t; in, This represents the predicted water output required for power generation at time t in the buffer zone; This represents the predicted outflow of water required to supply energy storage in the buffer zone at time t.

Images