An intelligent algorithm optimization combination and control method for parallel water pumps

Abstract

This invention proposes an intelligent algorithm optimization combination and control method for parallel water pumps. The method includes: configuring sensors and computing modules at the edge nodes of each water pump to collect and preprocess the operating data of each water pump in real time, generating a unified real-time feature vector and time series features; dynamically calculating the life state of the water pump based on the real-time feature vector and quantifying the remaining life; designing a dynamic start-stop scheme based on the life state based on a fuzzy logic algorithm to balance load demand and water pump life allocation; and optimizing the remaining life of the water pump from a global perspective based on federated learning, dynamically optimizing and adjusting the operating strategy based on the start-stop scheme. This invention constitutes a complete solution, providing a new technical path for the efficient and reliable operation of parallel water pumps.

Description

Technical Field

[0001] The present invention belongs to the field of parallel water pumps, and in particular relates to an intelligent algorithm optimization combination and control method for parallel water pumps. Background Art

[0002] Parallel pump systems are widely used in modern water supply networks, industrial circulating water systems, agricultural irrigation, and other applications due to their ability to flexibly adjust flow rates and adapt to diverse needs. By operating multiple pumps simultaneously, parallel pump systems dynamically adjust water supply capacity to accommodate load fluctuations and user demands. However, with increasing demand complexity and rising requirements for energy efficiency, lifespan, and stability, existing control technologies have exposed numerous operational issues, making them difficult to meet the demands of modern applications.

[0003] First, from the perspective of energy efficiency optimization, most current parallel pump control methods are based on simple operating curve control or fixed-rule start-stop methods. These methods typically operate according to the pump's design operating conditions or the equipment's start-stop sequence, making it difficult to optimize the pump combination in real time based on actual demand changes. Traditional control often aims to meet flow requirements, ignoring the efficiency characteristics and operating conditions of different pumps. This causes some pumps to operate under non-optimal operating conditions for extended periods of time, increasing overall energy consumption and potentially causing overloaded operation of some pumps. Furthermore, when the load fluctuates significantly, traditional methods experience a delayed response and frequent start-stop operations, which is not only inefficient but can also cause failures due to frequent operation.

[0004] Secondly, from the perspective of equipment life management, existing technologies for managing water pump life mostly remain at the static maintenance stage. For example, most systems only determine whether maintenance is needed based on the cumulative operating time or number of starts and stops of the equipment, and lack the dynamic adjustment capability of real-time life management. In actual operation, the life of a water pump is affected by a combination of factors, including the start-stop frequency, the fluctuation range of the operating pressure, the operating time, and the distribution of flow loads. The interaction of these factors will significantly accelerate the aging of the equipment. However, traditional methods do not take these dynamic factors into consideration, nor do they reasonably allocate the load and working time of the water pump according to the actual operating status, which can easily cause some equipment to fail prematurely, increase the maintenance cost of the system and the risk of unplanned downtime.

[0005] Third, from the perspective of system real-time performance and intelligence, most existing parallel pump control systems still rely on a centralized architecture. This centralized architecture relies on a central controller to collect and calculate the operating parameters of all pumps in order to unify the scheduling strategy. However, in large-scale parallel pump systems, this centralized model is prone to data transmission delays and computational bottlenecks. This makes it difficult for the central controller to issue optimization instructions in a timely manner, especially during periods of severe load fluctuations or emergency conditions, thus affecting system response speed. Furthermore, centralized systems have a high risk of single-point failures. Once a problem occurs with the central controller, the entire pump system may be paralyzed. Furthermore, traditional control methods remain at a relatively low level in terms of data analysis and model optimization, making it difficult to leverage real-time data for dynamic adaptation, and the overall intelligence level of the system is low.

[0006] In summary, existing parallel pump control methods have significant shortcomings in energy efficiency optimization, lifespan management, and real-time response capabilities. These technical deficiencies not only limit overall system efficiency but also increase equipment maintenance costs and reduce operational stability, seriously impacting the system's long-term reliability and economic benefits. Summary of the Invention

[0007] The purpose of this invention is to propose an intelligent algorithm optimization combination and control method for parallel water pumps, which can comprehensively optimize from system architecture to control strategy, taking into account energy efficiency, lifespan and real-time performance, and solving the bottleneck of existing technology.

[0008] In order to achieve the above object, the present invention provides a parallel water pump intelligent algorithm optimization combination and control method, the method comprising the following steps:

[0009] S1. Configure sensors and computing modules on the edge nodes of each water pump to collect and pre-process the operating data of each water pump in real time to generate a unified real-time feature vector and time series feature; the operating data includes the flow rate Q during the operation of the water pump. i , pressure P i , power consumption E i , start and stop times N i and the current load demand D;

[0010] The time series feature extraction is as follows:

[0011] Generate a feature sequence window in real time every minute Extract the historical trend of water pump operation, the length of which is the past 5 minutes of data, as shown below:

[0012]

[0013] Among them, X i [t-4] represents the feature vector of the window at time t-4, Xi [t-3] represents the feature vector of the window at time t-3, X i [t-2] represents the feature vector of the window at time t-2, X i [t-1] represents the feature vector of the window at time t-1, X i [t] represents the feature vector of the window at time t;

[0014] Each edge node will update the latest feature vector window every minute Upload to the central node for model training and global optimization;

[0015] S2. Dynamically calculate the life score of the water pump based on the real-time feature vector to quantify the remaining life;

[0016] S3. Designing a dynamic start-stop scheme based on a fuzzy logic algorithm according to the remaining life to balance load demand and pump life distribution;

[0017] S4: Optimize the remaining life of the water pump from a global perspective based on federated learning, and dynamically optimize and adjust the operation strategy based on the start-stop plan;

[0018] The design of a dynamic start-stop scheme based on a fuzzy logic algorithm according to the remaining life to balance load demand and pump life distribution specifically includes:

[0019] Design input variables and output variables for the fuzzy logic algorithm; wherein the input variables are life score, pressure fluctuation, and start-stop frequency, and the output variable is priority;

[0020] To meet the special needs of water pump start-stop optimization, a pressure-sensitive item and a start-stop wear compensation item are designed in combination with a fuzzy logic algorithm. The pressure-sensitive item is used to reduce the priority when the load fluctuates violently, preventing the water pump from excessively participating in the adjustment of unstable loads. The start-stop wear compensation item is used to dynamically adjust the priority and reduce the operating opportunities of water pumps with excessive start-stop frequencies. The pressure-sensitive item is expressed as follows:

[0021] ψ i =λ1·|ΔP i | 2

[0022] Where λ1 is the pressure fluctuation sensitivity factor, |ΔP i | 2 Used to amplify the suppressive effect of high pressure fluctuations on priority;

[0023] The start-stop wear compensation term is expressed as follows:

[0024] ξ i =λ2·max(0,ΔN i )

[0025] Among them, λ2 is the start-stop wear adjustment coefficient;

[0026] The input and output variables are combined with the pressure sensitivity term and the start-stop wear compensation term to express them as follows:

[0027] Priority i =μ L ·L i -ψ i -ξ i

[0028] Among them, L i is the lifespan score, μ L is the weight of the lifespan score, which directly reflects the main impact of lifespan on priority; ψ i It is a pressure-sensitive item that suppresses excessive burden on the water pump caused by fluctuations. i This is a start-stop wear compensation item that limits the continued operation of pumps with high start-stop frequencies;

[0029] Calculate priority i Finally, all water pumps are sorted by priority, and the load is distributed one by one from high to low until the current load demand D of the system is met; if the load demand drops to the lowest, the water pumps are stopped one by one from the lowest to the highest priority until the load meets the minimum requirement;

[0030] Finally, the optimization plan Schedule is generated.

[0031] Furthermore, the preprocessing includes:

[0032] For the flow Q i and flow rate P i For a continuous value signal, a sliding window smoothing algorithm is used to calculate the mean value every 5 sampling points to suppress random noise. The sliding window smoothing algorithm is expressed as follows:

[0033]

[0034] in, is the smoothed value at time t, w is the window size, X i [k] is the original sampling data;

[0035] The flow rate Q i , flow rate P i and energy consumption E i Normalized to [0,1] according to their respective historical extreme value ranges, expressed as follows:

[0036]

[0037] in, and The variables X are i The historical minimum and maximum values.

[0038] Furthermore, the real-time feature vector X i It is expressed as follows:

[0039]

[0040] in, Indicates the normalized value of flow; represents the normalized value of pressure; Represents the normalized value of power consumption; N i It indicates the number of starts and stops per unit time; D indicates the load demand.

[0041] Furthermore, the S2 specifically includes:

[0042] Constructing a water pump life status model Predict the life score L of the water pump based on the feature sequence window i , wherein the lifespan score L i The range is [0,1], 1 means the life is complete, 0 means the life is exhausted; the water pump life state model It is expressed as follows:

[0043]

[0044] Among them, β0 represents the initialization bias term, which represents the initial life of the device; β j Represents the weight of time series features, reflecting the contribution of different features to life prediction; γ is the cumulative fatigue factor weight coefficient; j represents time j; φ i Represents the cumulative fatigue factor, which describes the cumulative loss effect of high load and frequent start-stop;

[0045] When the water pump is actually running, the edge node of each water pump inputs its feature window in real time To life model And calculate the current life score L i , the life state is generated according to the calculation results, which is expressed as follows:

[0046]

[0047] in, Represents the time series feature window, providing short-term pump operation trend information; L i Represents the life score, which is used to quantify the remaining life of the current water pump;

[0048] A dynamic anomaly detection mechanism is designed to improve the robustness of the water pump life status model to abnormal data.

[0049] Furthermore, the cumulative fatigue factor is expressed as follows:

[0050]

[0051] Where ΔP i [k] is the pressure fluctuation amplitude at the kth moment in the window, N i [k] is the number of starts and stops, α k is the time decay weight.

[0052] Furthermore, the dynamic anomaly detection mechanism specifically includes:

[0053] If the difference between two consecutive calculated lifespan scores exceeds a threshold value δ, an abnormality flag is recorded and an automatic verification mechanism is triggered. The automatic verification mechanism specifically includes:

[0054] Complete Feature Window Upload to the central node.

[0055] The central node retrains using the latest collected pump operation status data The parameters β0,β j ,γ, avoids the error expansion caused by long-term drift of the model.

[0056] Furthermore, if the load demand is reduced to the minimum, the water pumps are deactivated one by one according to the priority from lowest to highest until the load meets the minimum requirement. The generated optimization plan is: Schedule = {Pump1: on, Pump2: on, Pump3: off}, and the plan update cycle is 1 minute.

[0057] Furthermore, the S4 specifically includes:

[0058] Design an optimization objective function J that comprehensively considers energy consumption and life loss; wherein the optimization objective function J satisfies the following constraints:

[0059] A. The total flow distribution meets the system load requirement D, which is expressed as follows:

[0060]

[0061] B. Flow distribution Q of each water pump i Not exceeding its maximum capacity, as indicated below:

[0062]

[0063] in, Indicates the maximum flow distribution of each pump;

[0064] C. The disabled pump does not distribute flow: If the pump i:off, then Q i =0;

[0065] The objective function J is optimized using the distributed gradient descent method;

[0066] After the optimization is completed, the optimal flow distribution plan FlowPlan={Q1,Q2,…,Q n}Satisfy the constraints and minimize the objective function J, and update the start-stop optimization plan Schedule ′ By checking the flow Q i Confirm: If Q i =0, then Pump i :off;

[0067] Collect optimized traffic distribution Q i and energy consumption data E i , as the calculation of the sample update life score to adapt it to the latest operating conditions.

[0068] Furthermore, the optimization objective function J is defined as follows:

[0069]

[0070] Where n represents the total number of water pumps, η i Indicates the flow rate ratio of pump i, E i represents energy consumption, σ i represents the life weighted term, L i represents the lifespan score;

[0071] The distributed gradient descent method is used to optimize the objective function J, specifically:

[0072] By priority i Sort the pumps and allocate the flow to the high-priority pumps first. The initial allocation satisfies the constraints and the gradient of the objective function J is calculated for each pump. The calculation is as follows:

[0073]

[0074] in, is the gradient of the effect of flow rate on life score;

[0075] At the same time, the gradient descent method is used to adjust the flow distribution of each water pump, as shown below:

[0076]

[0077] Among them, α is the step size factor, which ensures the stability of the optimization process;

[0078] When the flow adjustment of all water pumps meets the constraint conditions and the objective function change ΔJ is less than the set threshold, the iteration stops.

[0079] Furthermore, the flow rate ratio η of the water pump i is i , calculated as follows:

[0080]

[0081] Among them, Q i is the flow rate allocated to pump i, D is the total load demand of the system, is the maximum flow capacity of water pump i; the energy consumption E i It has a nonlinear relationship with the distribution flow rate and is calculated as follows:

[0082]

[0083] Among them, a i ,b i ,c i is the energy consumption coefficient of water pump i;

[0084] The life weighting term σ i It is used to dynamically reflect the risk level of water pump life consumption, and is calculated as follows:

[0085] σ i =1+α·exp(-β·L i )

[0086] Among them, α and β are adjustment coefficients used to control the extent of risk increase; λ is the life protection weight, which balances energy consumption optimization and life optimization; when L i When the minimum i If the maximum value is reached, the flow distribution of the current water pump i will be reduced to protect its life.

[0087] The beneficial technical effects of the present invention are at least as follows:

[0088] This invention utilizes edge computing technology to transform traditional centralized control into a distributed architecture. Independent edge nodes are deployed on each pump, responsible for local data collection and preliminary calculations, and dynamically adjust operating parameters under the guidance of a central node. This distributed architecture significantly reduces data transmission latency, improves system response speed, and enhances system robustness and fault tolerance, resolving the real-time limitations and single-point failure risks of traditional centralized control.

[0089] To address the shortcomings of water pump lifespan management, this invention incorporates a fuzzy logic algorithm to dynamically adjust the start / stop priority and operating frequency of water pumps based on real-time data such as pressure fluctuations, start / stop frequency, and current lifespan status. Fuzzy control effectively balances the load across pumps, preventing overuse of certain equipment and thus extending the overall system lifespan. This addresses the inability of traditional methods to dynamically adapt lifespan management.

[0090] This paper proposes a dual-objective optimization algorithm that comprehensively considers the energy consumption characteristics and lifespan of water pumps. Through real-time load distribution adjustments, it achieves a synergistic improvement in energy efficiency optimization and lifespan balance. In the short term, this algorithm uses edge nodes to achieve real-time energy consumption optimization. In the long term, it uses federated learning to integrate global data through central nodes to further optimize overall system performance, resolving the issue of separating energy efficiency optimization from lifespan management.

[0091] This invention achieves a comprehensive breakthrough from system architecture to control algorithm by combining edge computing and fuzzy control. Edge computing improves real-time performance and reliability, while fuzzy control provides dynamic lifespan optimization capabilities. Dual-objective optimization enables the coordinated management of energy efficiency and lifespan. These complementary innovations form a complete solution, providing a new technical path for the efficient and reliable operation of parallel water pumps. BRIEF DESCRIPTION OF THE DRAWINGS

[0092] The present invention is further described with reference to the accompanying drawings. However, the embodiments in the accompanying drawings do not constitute any limitation to the present invention. A person skilled in the art can obtain other drawings based on the following drawings without creative effort.

[0093] Figure 1 This is a flow chart of the intelligent algorithm optimization combination and control method of parallel water pumps of the present invention. DETAILED DESCRIPTION

[0094] The following describes embodiments of the present invention in detail. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended only to explain the present invention and are not to be construed as limiting the present invention.

[0095] like Figure 1 As shown, an embodiment of the present invention provides an intelligent algorithm optimization combination and control method for parallel water pumps, the method comprising:

[0096] S1. Configure sensors and computing modules on the edge nodes of each water pump to collect and pre-process the operating data of each water pump in real time to generate a unified real-time feature vector and time series feature; the operating data includes the flow rate Q during the operation of the water pump. i , pressure Pi , power consumption E i , start and stop times N i and the current load demand D.

[0097] Specifically, sensors and computing modules are configured on the edge nodes of each water pump to collect real-time data including the flow rate (Q i , unit: cubic meter / second), pressure (P i , unit: Pascal), power consumption (E i , unit: Watt), start and stop times (N i , unit: times) and current load demand (D, unit: cubic meters / second).

[0098] Among them, the flow Q i The pressure is measured accurately by ultrasonic or electromagnetic flowmeter, sampling once per second. i The pressure value of the water pump outlet is collected through the pressure sensor, and the sample is taken once per second. i The real-time power consumption of the pump is calculated through the power metering module. i Based on the accumulation of pump start and stop signals (high level / low level), the start and stop frequency per minute is recorded. The current load demand D is provided by the system demand forecast module and is updated every minute.

[0099] Furthermore, based on the characteristics of data from different sources, the following strategies are used for preprocessing:

[0100] Denoising: For Q i and P i For signals with equal continuous values, a sliding window smoothing algorithm is used, and the mean is calculated every 5 sampling points to suppress random noise. The formula is as follows:

[0101]

[0102] in, is the smoothed value at time t, w is the window size (value 5), X i [k] is the original sampling data.

[0103] Normalization: In order to facilitate subsequent life evaluation, Q i 、P i and E i Normalized to [0,1] according to their respective historical extreme value ranges:

[0104]

[0105] in, and The variables X are i The historical minimum and maximum values.

[0106] The preprocessed variables are constructed into the real-time feature vector of the water pump:

[0107]

[0108] in, Indicates the normalized value of the flow rate. Indicates the normalized value of pressure. Indicates the normalized value of power consumption. N i It represents the number of starts and stops per unit time and does not need to be normalized. D represents the load demand of the system and is directly used as input.

[0109] Furthermore, in order to extract the historical trend of water pump operation, a feature sequence window is generated in real time every minute, with a length of the past 5 minutes of data. The form of each feature vector is:

[0110]

[0111] This window is used to capture short-term operational changes and provide input to the life assessment module.

[0112] Furthermore, edge node upload rules:

[0113] Each edge node will update the latest feature vector window every minute Upload to the central node for model training and global optimization. When the set threshold (0.8) is exceeded, the high-frequency upload mechanism is triggered and all historical feature data are sent to the central node.

[0114] The high-frequency upload mechanism specifically includes:

[0115] Edge nodes establish communication connections with central nodes via wired or wireless means. Under high-frequency upload mechanisms, this connection may require higher bandwidth and lower latency to ensure real-time data transmission.

[0116] Once the connection is established, the edge node will send data to the central node at a higher frequency. This high-frequency transmission ensures that the data can reach the central node quickly in abnormal situations for subsequent analysis and processing;

[0117] After receiving the data, the central node will confirm and verify it to ensure the integrity and accuracy of the data; if the data is incorrect or lost, the central node may send a feedback message to the edge node, requesting it to resend or supplement the data.

[0118] S2. Dynamically calculate the life score of the water pump based on the real-time feature vector and quantify the remaining life.

[0119] Specifically, define a water pump life state model The model predicts the life score L of the water pump based on the feature sequence window i (The range is [0,1], 1 means the life is complete, 0 means the life is exhausted.) The model design combines the start-stop loss of the pump mechanical components with the long-term impact of load fluctuations. The formula is as follows:

[0120]

[0121] Among them, β0 represents the initialization bias term, which represents the initial life of the equipment (such as L for a new water pump). i =1). β j Represents the weight of the time series feature, reflecting the contribution of different features to life expectancy prediction, obtained by model training. i represents the cumulative fatigue factor, which describes the cumulative loss effect of high load and frequent start-stop, and is defined as:

[0122]

[0123] Where ΔP i [k] is the pressure fluctuation amplitude at the kth moment in the window (unit: Pascal), N i [k] is the number of starts and stops, α k is the time decay weight (e.g. α k =0.8 k-1 ).

[0124] In the actual operation of the system, the edge node of each water pump inputs its feature window in real time To life model And calculate the current life score L i The life state is generated based on the calculation results:

[0125]

[0126] Lifecycle states include: Represents the time series feature window, providing short-term pump operation trend information. i Represents the life score, which is used to quantify the remaining life of the current water pump.

[0127] Furthermore, to improve the model's robustness to abnormal data, a dynamic anomaly detection mechanism was designed:

[0128] If the difference between two consecutive calculated lifespan scores exceeds a threshold value δ (eg, δ=0.1), an abnormal flag is recorded and an automatic verification mechanism is triggered.

[0129] The automatic verification process includes:

[0130] Complete Feature Window Upload to the central node. ;

[0131] The central node is retrained using the latest data The parameters β0,β j ,γ, avoids the error expansion caused by long-term drift of the model.

[0132] Furthermore, the edge node updates the lifespan state every minute. i , and serves as a direct input for the next step of pump start-stop optimization to control the priority and scheduling strategy of the pump.

[0133] S3. Design a dynamic start-stop plan based on a fuzzy logic algorithm according to the remaining life to balance the load demand and the life distribution of the pump.

[0134] Specifically, the fuzzy control design is based on the actual working conditions of the water pump system, selects three input variables, and calculates the output variable priority i :

[0135] Input variables:

[0136] Lifespan score L i : The current life status of the water pump, L i The lower the value, the lower the priority.

[0137] pressure fluctuations The difference between the current pressure and the average pressure in the window indicates the impact of load fluctuations on the operation of the pump.

[0138] Start and stop frequency The difference between the current start and stop times and the historical average value within the window is used to reflect the wear status of mechanical components.

[0139] Output variables:

[0140] Priority i : The range is [0,1], indicating the start and stop priority of the water pump. The higher the value, the higher the priority.

[0141] Design pressure sensitivity and start-stop wear compensation

[0142] In response to the special needs of water pump start-stop optimization, the following innovative items are designed to enhance the adaptability of the model:

[0143] Pressure sensitivity term ψ i : Used to reduce the priority when the load fluctuates violently, to prevent the pump from excessively participating in the adjustment of unstable loads:

[0144] ψ i =λ1·|ΔP i | 2

[0145] Where λ1 is the pressure fluctuation sensitivity factor (typical value λ1 = 0.1), |ΔP i | 2 Used to amplify the suppressive effect of high pressure fluctuations on priority.

[0146] Start-stop wear compensation item ξ i : Used to dynamically adjust the priority and reduce the running opportunities of pumps with too high start and stop frequency:

[0147] ξ i =λ2·max(0,ΔN i )

[0148] Wherein, λ2 is the start-stop wear adjustment coefficient (typical value λ2 = 0.05).

[0149] Furthermore, considering the above factors, the priority i Calculated by the following formula:

[0150] Priority i =μ L ·L i -ψ i -ξ i

[0151] Among them, μ L is the weight of the life score (typical value μ L =1.0), directly reflecting the main impact of lifespan on priority. i It is a pressure-sensitive item that suppresses excessive burden on the water pump caused by fluctuations. i To compensate for the start and stop wear, the pump with a high start and stop frequency is restricted from continuing to run.

[0152] Furthermore, the priority is calculated i Finally, all pumps are sorted by priority, and the load is distributed one by one from high to low until the current system load demand D (unit: cubic meters / second) is met. If the system load demand decreases, the pumps are deactivated one by one from low to high priority until the load meets the minimum requirement.

[0153] It can be understood. For example: Assume that the system currently has 3 water pumps, whose life scores are [0.9, 0.7, 0.6], and the pressure fluctuation ΔP i is [0.05, 0.1, 0.02], the start-stop frequency ΔN i is [0, 2, 1], and the current load demand D = 70 (unit: cubic meters / second).

[0154] Calculate the priority of each pump:

[0155] Priority1=μL ·0.9-ψ1-ξ1=1.0·0.9-0.1·(0.05) 2 -0=0.89975.

[0156] Priority2=μ L ·0.7-ψ2-ξ2=1.0·0.7-0.1·(0.1) 2 -0.05·2=0.685.

[0157] Priority3=μ L ·0.6-ψ3-ξ3=1.0·0.6-0.1·(0.02) 2 -0.05·1=0.59596.

[0158] Sort by Priority i =[Pump1:0.89975,Pump2:0.685,Pump3:0.59596]. When distributing the load, the first two pumps are started first to meet the load requirement of D=70, and the third pump is stopped.

[0159] Furthermore, the generated optimization plan is: Schedule = {Pump1: on, Pump2: on, Pump3: off}, and the plan update cycle is 1 minute.

[0160] S4. Based on federated learning, optimize the remaining life of the water pump from a global perspective and dynamically optimize and adjust the operation strategy based on the start-stop plan.

[0161] Specifically, in order to reflect the characteristics of the patent, an optimization objective function that comprehensively considers energy consumption and life loss is designed and defined as follows:

[0162]

[0163] Where n represents the total number of water pumps. i It represents the flow rate ratio of pump i, which is calculated as follows:

[0164]

[0165] Among them, Q i is the flow rate allocated to pump i (unit: cubic meters / second), D is the total load demand of the system, is the maximum flow capacity of pump i.

[0166] E i The energy consumption of pump i is expressed in watts (W), which has a nonlinear relationship with the distribution flow rate and is usually fitted by an empirical formula:

[0167]

[0168] Among them, a i ,b i ,c i is the energy consumption coefficient of pump i, which depends on its design characteristics. i represents the life score of pump i, which is obtained from the life model in step 2. Updated in real time. i It represents the innovative life weighted item, dynamically reflecting the risk level of the pump life consumption, and is defined as:

[0169] σ i =1+α·exp(-β·L i )

[0170] Understandably, when L i When σ is low, i The rapid increase forces the optimization algorithm to reduce the flow distribution of the pump to protect its life.

[0171] Where α and β are adjustment coefficients (e.g., α = 2, β = 5) used to control the magnitude of risk increase. λ represents the life protection weight, balancing energy consumption optimization with life optimization, with a typical value of λ∈[0.1,1.0].

[0172] Furthermore, the optimization problem needs to satisfy the following constraints:

[0173] The total flow distribution meets the system load requirements:

[0174]

[0175] The flow distribution of each pump does not exceed its maximum capacity:

[0176]

[0177] The disabled pump does not distribute flow: If the Pump i :off, then Q i =0.

[0178] Furthermore, the distributed gradient descent method is used to optimize the objective function J. The algorithm steps are as follows:

[0179] Initialization: by priority i Sort the pumps and allocate flow to high-priority pumps first, and the initial allocation satisfies the constraints.

[0180] Gradient calculation: Calculate the gradient of each pump for the objective function J

[0181]

[0182] in, is the gradient of the effect of flow on life score, which is determined by the life model The parameters in are calculated.

[0183] Update flow distribution: Use gradient descent to adjust the flow distribution of each pump:

[0184]

[0185] Wherein, α is the step size factor to ensure the stability of the optimization process (e.g., α = 0.01).

[0186] Iteration termination condition: When the flow adjustment of all pumps meets the constraint conditions and the objective function change ΔJ is less than the set threshold (such as ΔJ < 10 -3 ), stop the iteration.

[0187] Furthermore, after the optimization is completed, the optimal flow distribution plan FlowPlan={Q1,Q2,…,Q n}Satisfy the constraints while minimizing the objective function J.

[0188] Updated start-stop optimization plan Schedule ′ By checking the flow Q i Confirm: If Q i =0, then Pump i :off.

[0189] Example: Assume that the system has 3 water pumps and their initial states are as follows:

[0190] Lifespan score L i =[0.9,0.7,0.5], maximum flow The current load demand D=70.

[0191] Initial flow distribution: Q1=40, Q2=30, Q3=0.

[0192] Through iterative optimization, the final flow distribution result is FlowPlan = {35,35,0}, corresponding to the start-stop plan Schedule ′ ={Pump1:on,Pump2:on,Pump3:off}.

[0193] Furthermore, the optimized traffic distribution Q is collected i and energy consumption data E i , as a sample to update the life model The central node periodically integrates data from all edge nodes, uses federated learning technology to update the global optimization model, and sends it to each edge node to achieve long-term performance improvement.

[0194] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0195] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is merely a division of logical functions. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0196] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0197] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the claims and their equivalents.

Claims

1. A parallel water pump intelligent algorithm optimization combination and control method, characterized in that: The method comprises the following steps: S1. Configure sensors and computing modules on the edge nodes of each water pump to collect and pre-process the operating data of each water pump in real time to generate a unified real-time feature vector and time series feature; the operating data includes the flow rate Q during the operation of the water pump. i , pressure P i , power consumption E i , start and stop times N i and the current load demand D; The time series feature extraction is as follows: Generate a feature sequence window in real time every minute Extract the historical trend of water pump operation, the length of which is the past 5 minutes of data, as shown below: Among them, X i [t-4] represents the feature vector of the window at time t-4, X i [t-3] represents the feature vector of the window at time t-3, X i [t-2] represents the feature vector of the window at time t-2, X i [t-1] represents the feature vector of the window at time t-1, X i [t] represents the feature vector of the window at time t; The edge node upload rules are as follows: Each edge node will update the latest feature vector window every minute Upload to the central node for model training and global optimization; however, when abnormal fluctuations are detected, each edge node triggers a high-frequency upload mechanism to send all historical feature data to the central node; S2. Dynamically calculate the life status of the water pump based on the real-time characteristic vector and quantify the remaining life; S3. Designing a dynamic start-stop scheme based on a fuzzy logic algorithm according to the life status to balance load demand and pump life distribution; S4: Optimize the remaining life of the water pump from a global perspective based on federated learning, and dynamically optimize and adjust the operation strategy based on the start-stop plan; The design of a dynamic start-stop scheme based on a fuzzy logic algorithm according to the life state to balance load demand and pump life distribution specifically includes: Design input variables and output variables for the fuzzy logic algorithm; wherein the input variables are life score, pressure fluctuation, and start-stop frequency, and the output variable is priority; To meet the special needs of water pump start-stop optimization, a pressure-sensitive item and a start-stop wear compensation item are designed in combination with a fuzzy logic algorithm. The pressure-sensitive item is used to reduce the priority when the load fluctuates violently, preventing the water pump from excessively participating in the adjustment of unstable loads. The start-stop wear compensation item is used to dynamically adjust the priority and reduce the operating opportunities of water pumps with excessive start-stop frequencies. The pressure-sensitive item is expressed as follows: ψ i =λ1·|ΔP i | 2 Where λ1 is the pressure fluctuation sensitivity factor, |ΔP i | 2 Used to amplify the suppressive effect of high pressure fluctuations on priority; The start-stop wear compensation term is expressed as follows: x i =λ2·max(0,ΔN i ) Among them, λ2 is the start-stop wear adjustment coefficient; The input and output variables are combined with the pressure sensitivity term and the start-stop wear compensation term to express them as follows: Priority i =μ L ·L i -ψ i -x i Among them, L i is the lifespan score, μ L is the weight of the lifespan score, which directly reflects the main impact of lifespan on priority; ψ i It is a pressure-sensitive item that suppresses excessive burden of fluctuation on the water pump; i This is a start-stop wear compensation item that limits the continued operation of pumps with high start-stop frequencies; Calculate Priority i Finally, all water pumps are sorted by priority, and the load is distributed one by one from high to low until the current load demand D of the system is met; if the load demand drops to the lowest, the water pumps are stopped one by one from the lowest to the highest priority until the load meets the minimum requirement; Finally, generate the optimization plan Schedule; Wherein, the S2 specifically includes: Constructing a water pump life status model Predict the life score L of the water pump based on the feature sequence window i , wherein the lifespan score L i The range is [0,1], 1 means the life is complete, 0 means the life is exhausted; the water pump life state model It is expressed as follows: Among them, β0 represents the initialization bias term, which represents the initial life of the device; β j Represents the weight of time series features, reflecting the contribution of different features to life prediction; γ is the cumulative fatigue factor weight coefficient; j represents time j; φ i Represents the cumulative fatigue factor, which describes the cumulative loss effect of high load and frequent start-stop; When the water pump is actually running, the edge node of each water pump inputs its feature window in real time To life model And calculate the current life score L i , generate life state State according to the calculation results i , which is expressed as follows: in, Represents the time series feature window, providing short-term pump operation trend information; L i Represents the life score, which is used to quantify the remaining life of the current water pump; Design a dynamic anomaly detection mechanism to improve the robustness of the pump life state model to abnormal data; The dynamic anomaly detection mechanism specifically includes: If the difference between two consecutive calculated lifespan scores exceeds a threshold value δ, an abnormality flag is recorded and an automatic verification mechanism is triggered. The automatic verification mechanism specifically includes: Complete Feature Window Upload to the central node; The central node retrains using the latest collected pump operation status data The parameters β0,β j ,γ, avoids the error expansion caused by long-term drift of the model; Wherein, the S4 specifically includes: Design an optimization objective function J that comprehensively considers energy consumption and life loss; the optimization objective function J is defined as follows: Where n represents the total number of water pumps, η i Indicates the flow rate ratio of pump i, E i represents energy consumption, σ i represents the life weighted term, L i represents the lifespan score; The optimization objective function J satisfies the following constraints: A. The total flow distribution meets the system load requirement D, which is expressed as follows: B. Flow distribution Q of each water pump i Not exceeding its maximum capacity, as indicated below: in, Indicates the maximum flow distribution of each pump; C. The deactivated pump does not distribute flow: If the pump i :off, then Q i =0; The objective function J is optimized using the distributed gradient descent method; After the optimization is completed, the optimal flow distribution plan FlowPlan={Q1,Q2,…,Q n }Satisfy the constraints and minimize the objective function J, and update the start-stop optimization plan Schedule ′ By checking the flow Q i Confirm: If Q i =0, then Pump i :off; Collect optimized traffic distribution Q i and energy consumption data E i , as the calculation of the sample update life score to adapt it to the latest operating conditions.

2. The intelligent algorithm optimization combination and control method for parallel water pumps according to claim 1 is characterized in that: The pretreatment includes: For the flow Q i and flow rate P i For a continuous value signal, a sliding window smoothing algorithm is used to calculate the mean value every 5 sampling points to suppress random noise. The sliding window smoothing algorithm is expressed as follows: in, is the smoothed value at time t, w is the window size, X i [k] is the original sampling data; The flow rate Q i , flow rate P i and energy consumption E i Normalized to [0,1] according to their respective historical extreme value ranges, expressed as follows: in, and The variables X are i The historical minimum and maximum values.

3. The intelligent algorithm optimization combination and control method for parallel water pumps according to claim 2 is characterized in that: The real-time feature vector X i It is expressed as follows: in, Indicates the normalized value of flow; represents the normalized value of pressure; Represents the normalized value of power consumption; N i It indicates the number of starts and stops per unit time; D indicates the load demand.

4. The intelligent algorithm optimization combination and control method for parallel water pumps according to claim 1 is characterized in that: The cumulative fatigue factor is expressed as follows: Where ΔP i [k] is the pressure fluctuation amplitude at the kth moment in the window, N i [k] is the number of starts and stops, α k is the time decay weight.

5. The intelligent algorithm optimization combination and control method for parallel water pumps according to claim 1 is characterized in that: If the load demand is reduced to the minimum, the water pumps are stopped one by one from the lowest to the highest priority until the load meets the minimum requirement. The generated optimization plan is: Schedule = {Pump1: on, Pump2: on, Pump3: off}, and the plan update cycle is 1 minute.

6. The intelligent algorithm optimization combination and control method for parallel water pumps according to claim 1 is characterized in that: The distributed gradient descent method is used to optimize the objective function J, specifically: By priority i Sort the pumps and allocate the flow to the high-priority pumps first. The initial allocation satisfies the constraints and the gradient of the objective function J is calculated for each pump. The calculation is as follows: in, is the gradient of the effect of flow rate on life score; At the same time, the gradient descent method is used to adjust the flow distribution of each water pump, as shown below: Among them, α is the step size factor, which ensures the stability of the optimization process; When the flow adjustment of all water pumps meets the constraint conditions and the objective function change ΔJ is less than the set threshold, the iteration stops.

7. The intelligent algorithm optimization combination and control method for parallel water pumps according to claim 6 is characterized in that: The flow rate ratio η of the water pump i i , calculated as follows: Among them, Q i is the flow rate allocated to pump i, D is the total load demand of the system, is the maximum flow capacity of pump i; The energy consumption E i It has a nonlinear relationship with the distribution flow rate and is calculated as follows: Among them, a i ,b i ,c i is the energy consumption coefficient of water pump i; The life weighting term σ i It is used to dynamically reflect the risk level of water pump life consumption, and is calculated as follows: s i =1+α·exp(-β·L i ) Among them, α and β are adjustment coefficients used to control the extent of risk increase; λ is the life protection weight, which balances energy consumption optimization and life optimization; when L i When the minimum i If the maximum value is reached, the flow distribution of the current water pump i will be reduced to protect its life.

Images