A pump set energy consumption optimization method based on machine learning

Abstract

The present application relates to pump group energy consumption optimization technical field, specifically disclose a kind of pump group energy consumption optimization method based on machine learning, comprising the following steps: step S1: obtaining all the pump group jurisdiction and the electric valve on the pipe line, for each electric valve independently generating opening micro-vibration signal, signal amplitude is set as preset micro-vibration amplitude, and the signal frequency of each electric valve is not mutually different with the rest valve;Step S2: real-time acquisition pump group total input power and the real-time flow of each pipe line, for each pipe line generates sinusoidal reference signal, the power variation component corresponding pipe line is separated from total power by phase-sensitive demodulation operation, and accordingly generates reverse opening correction amount;While comparing pipe line real-time flow and demand flow, the correction amount is adjusted;Step S3: real-time check the opening correction amount of each pipe line after adjustment, if the value exceeds preset convergence threshold, then adjust corresponding electric valve opening according to correction amount.

Description

Technical Field

[0001] This invention relates to the field of pump set energy consumption optimization technology, and specifically to a pump set energy consumption optimization method based on machine learning. Background Technology

[0002] Pre-integrated cooling plants have widely adopted a multi-unit cooling tower bidirectional or unidirectional flow multi-pump backup distribution architecture. By connecting multiple pumps in parallel and using them as backups for each other, the reliability of system operation is improved. Relying on the frequency conversion regulation of water pumps, the cooling capacity distribution can be adapted under variable flow conditions to meet the dynamic cooling needs of building terminals.

[0003] However, conventional integrated chiller plants only achieve equipment-level integration and packaging. The internal distribution pipelines are not systematically optimized according to a low-resistance architecture, resulting in high local fluid resistance that fluctuates dynamically with operating conditions and equipment aging. Existing control methods rely on static parameters and fixed setpoints, failing to perceive the actual impedance distribution of the pipeline network online. This leads to poor adaptability under variable flow conditions, easily causing hydraulic imbalances such as near-end overflow and far-end underflow, generating significant ineffective throttling losses. Consequently, the proportion of pump energy consumption in the total system energy consumption remains consistently high, preventing the true maximization of energy efficiency in integrated chiller plant systems. Summary of the Invention

[0004] The purpose of this invention is to provide a pump group energy consumption optimization method based on machine learning, and to solve the following technical problems.

[0005] The objective of this invention can be achieved through the following technical solutions: A machine learning-based method for optimizing pump group energy consumption includes the following steps: Step S1: Obtain all the transmission and distribution pipelines of the pump group, and obtain the electric valves of each transmission and distribution pipeline. For any electric valve, generate an opening micro-vibration signal for the electric valve. The amplitude of the opening micro-vibration signal is a preset micro-vibration amplitude value, and the frequency is different from that of the other electric valves. Step S2: Obtain the real-time total input power of the pump set. For any distribution pipeline, obtain the real-time flow rate of the distribution pipeline and generate a sinusoidal reference signal with the same frequency as the opening micro-vibration signal of the distribution pipeline. Perform phase-sensitive demodulation operation on the sinusoidal reference signal and the real-time total input power to obtain the pump set power variation component of the distribution pipeline. The pump set power variation component includes a direction component and an amplitude component. Based on the directional component, an opening correction amount is generated with a direction opposite to the directional component and an amplitude of a preset opening amplitude threshold; the demand value of the transmission and distribution pipeline is obtained, the real-time flow rate is compared with the demand value, and the opening correction amount is adjusted according to the comparison result to obtain the adjusted opening correction amount; Step S3: For any distribution pipeline, determine in real time whether the adjusted opening correction amount exceeds the preset convergence threshold. If it exceeds the preset convergence threshold, adjust the opening of the electric valve of the distribution pipeline according to the adjusted opening correction amount.

[0006] As a further aspect of the present invention: the generation process of the micro-vibration signal for the opening degree of each electric valve includes: A fundamental frequency is set, and the frequencies of the micro-vibration signals for the opening degree of each electric valve are configured to be integer multiples of the fundamental frequency, with the integer multiples for each electric valve being coprime; the amplitude when the electric valve is fully open is obtained and denoted as the maximum amplitude F. max The preset micro-amplitude value is configured to be within the range of [1%F]. max 5%F max ].

[0007] As a further aspect of the present invention: the process of performing phase-sensitive demodulation operation on the sinusoidal reference signal and the real-time total input power includes: The real-time total input power is multiplied by the sinusoidal reference signal to obtain a product signal; the product signal is low-pass filtered to remove components with frequencies exceeding the preset micro-amplitude value to obtain a DC component; the sign of the DC component is obtained and recorded as the direction component; the amplitude of the DC component is obtained and recorded as the amplitude component; the pump power variation component is composed of the direction component and the amplitude component.

[0008] As a further aspect of the present invention: the process of determining the amplitude of the opening correction includes: Obtain the amplitude components of each transmission and distribution pipeline, select the amplitude component with the largest value, and record it as the maximum amplitude component; for any transmission and distribution pipeline, obtain the ratio of the amplitude component of the transmission and distribution pipeline to the maximum amplitude component, multiply the ratio by the preset micro-vibration amplitude value to obtain the correction amplitude, and set the correction amplitude as the opening amplitude threshold.

[0009] As a further aspect of the present invention: the process of adjusting the opening correction amount based on the comparison results includes: If the real-time flow is less than the demand value, the direction of the opening correction amount is adjusted, and the opening correction amount after the direction adjustment is recorded as the adjusted opening correction amount; if the real-time flow is greater than or equal to the demand value, the opening correction amount is retained and directly recorded as the adjusted opening correction amount.

[0010] As a further aspect of the present invention: the process of adjusting the direction of the opening correction amount includes: If the direction of the opening correction amount is the direction of decreasing opening, then the direction of the opening correction amount is adjusted to the direction of increasing opening; if the direction of the opening correction amount is the direction of increasing opening or is zero, then it remains unchanged.

[0011] As a further aspect of the present invention, the process of obtaining the adjusted opening correction amount further includes: The amplitude of the adjusted opening correction is compared with the preset maximum step size for a single cycle. If the amplitude of the adjusted opening correction is greater than the maximum step size for a single cycle, the amplitude of the adjusted opening correction is truncated to the maximum step size for a single cycle, while the direction remains unchanged.

[0012] As a further aspect of the present invention: the process of adjusting the opening degree of the electric valve in the transmission and distribution pipeline includes: The system obtains the current real-time total flow rate and real-time total head of the pump set. When the real-time total head is lower than the preset minimum safe head, the power supply frequency of the pump set is increased. When the real-time total flow rate exceeds a preset proportion of the sum of the demand values ​​of each transmission and distribution pipeline, the power supply frequency of the pump set is decreased.

[0013] The beneficial effects of this invention are: This invention achieves distributed online identification of the impact of impedance changes in each distribution pipeline on the power of the pump unit by injecting micro-amplitude opening disturbance signals with different frequencies into the electric valves of each distribution pipeline and performing phase-sensitive demodulation calculation on the total input power of the pump unit. This invention eliminates the need for prior acquisition of any system parameters such as pipeline topology, pipeline length, and pipe diameter, and also eliminates the need for establishing complex hydraulic models or training neural networks. It relies solely on the correlation extraction between the globally measurable signal of the total pump power and the disturbance signals of each valve to isolate the power gradient direction for each valve, significantly reducing the deployment threshold and computing power requirements. During the optimization process, the opening correction of the electric valves in each distribution pipeline is entirely based on locally available power gradient estimates and the demand value of the terminal flow. Furthermore, this invention introduces the constraint of terminal flow demand on the opening correction direction. The mechanism of this invention, while pursuing a reduction in pump power, strictly ensures the minimum cooling flow rate at each terminal, fundamentally solving the hydraulic imbalance problem of near-end overflow and far-end insufficient flow under traditional constant differential pressure control, as well as the resulting ineffective throttling losses; and through continuous trial, gradient estimation and closed-loop iteration, the electric valves of each distribution pipeline are gradually guided to the opening combination that meets the flow requirements of all terminals and the total power of the pump unit converges to a minimum value, automatically entering the optimal energy-saving condition of on-demand flow distribution and pump head only overcoming the necessary pipeline resistance without any external setting value. Attached Figure Description

[0014] The invention will now be further described with reference to the accompanying drawings.

[0015] Figure 1 This is a schematic diagram illustrating the steps of a pump group energy consumption optimization method based on machine learning according to the present invention. Figure 2This is a schematic diagram of the structure of a pump group energy consumption optimization method based on machine learning according to the present invention. Detailed Implementation

[0016] 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 embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Understandably, this invention injects a tiny, frequency-different identification tag into each branch valve, namely an opening micro-vibration signal. The micro-vibration of the opening micro-vibration signal causes a tiny fluctuation in the total power of the pump group. Through phase-sensitive demodulation calculation, the direction of power fluctuation caused by each electric valve individually is decoded from the total power signal, thereby determining whether the electric valve is opening wide to save energy or closing narrow to save energy. Then, the valve opening is gradually adjusted along the energy-saving direction, while directional protection is provided for branches with insufficient flow. Finally, under the premise of meeting the flow requirements at the end, the system automatically converges to the valve opening combination with the lowest energy consumption of the pump group.

[0018] Please see Figure 1 As shown, this invention is a pump group energy consumption optimization method based on machine learning, including the following steps: Step S1: Obtain all the transmission and distribution pipelines of the pump group, and obtain the electric valves of each transmission and distribution pipeline. For any electric valve, generate an opening micro-vibration signal for the electric valve. The amplitude of the opening micro-vibration signal is a preset micro-vibration amplitude value, and the frequency is different from that of the other electric valves. It should be noted that this invention is applied to a multi-unit cooling tower bidirectional or unidirectional flow multi-pump low-resistivity distribution pump system in an integrated cooling plant. The distribution system consists of a shared distribution main pipe and parallel-connected branch distribution pipes. Each branch distribution pipe is equipped with an electric valve for flow regulation. The system traverses and acquires the equipment node information of all distribution pipes and corresponding electric valves, and independently generates an opening micro-vibration signal for each electric valve. The opening micro-vibration signal is based on a preset reference frequency, and the signal frequency corresponding to each electric valve is configured according to mutually distinct coprime integer multiples. At the same time, the amplitude of the opening micro-vibration signal is set to a preset micro-vibration amplitude value. The generated opening micro-vibration signal is superimposed on the current steady-state opening command of the corresponding electric valve, so that each electric valve generates continuous micro-amplitude fluctuations on the basis of normal operating opening. Moreover, the fluctuation characteristics of each electric valve are independent of each other in the frequency domain. This does not cause substantial disturbance to the steady-state fluid characteristics of the low-resistivity distribution architecture, and provides independent and separable feature identifiers for subsequent power gradient identification. In a preferred embodiment of the present invention, the generation process of the opening micro-vibration signal of each electric valve includes: A fundamental frequency is set, and the frequencies of the micro-vibration signals for the opening degree of each electric valve are configured to be integer multiples of the fundamental frequency, with the integer multiples for each electric valve being coprime; the amplitude when the electric valve is fully open is obtained and denoted as the maximum amplitude F. max The preset micro-amplitude value is configured to be within the range of [1%F]. max 5%F max ]; It is understandable that the frequency configuration is set to an integer multiple of the coprime fundamental frequency to ensure that the micro-vibration signals of each distribution pipeline are strictly orthogonal in the frequency domain. If the frequencies of two valves have an integer multiple harmonic relationship, they will crosstalk each other during demodulation, leading to incorrect gradient estimation. Using an integer multiple of the coprime fundamental frequency can avoid this spectral aliasing to the greatest extent. Set the range to [1%F] max 5%F max As a compromise in engineering, if the amplitude is too small, the pump power fluctuation will be drowned out by the measurement noise and cannot be effectively demodulated; if the amplitude is too large, it will cause a perceptible disturbance to the flow rate at the end, affecting user comfort. Step S2: Obtain the real-time total input power of the pump set. For any distribution pipeline, obtain the real-time flow rate of the distribution pipeline and generate a sinusoidal reference signal with the same frequency as the opening micro-vibration signal of the distribution pipeline. Perform phase-sensitive demodulation operation on the sinusoidal reference signal and the real-time total input power to obtain the pump set power variation component of the distribution pipeline. The pump set power variation component includes a direction component and an amplitude component. Based on the directional component, an opening correction amount is generated with a direction opposite to the directional component and an amplitude of a preset opening amplitude threshold; the demand value of the transmission and distribution pipeline is obtained, the real-time flow rate is compared with the demand value, and the opening correction amount is adjusted according to the comparison result to obtain the adjusted opening correction amount; It should be noted that this invention relies on the global acquisition parameters of a multi-pump shared low-resistance distribution system to collect the total input power of the pump group's main power supply circuit and the real-time flow rate on each branch's distribution pipeline. For each distribution pipeline, a sinusoidal reference signal with the same frequency as the opening micro-vibration signal of the electric valve of the distribution pipeline is generated. The sinusoidal reference signal is multiplied by the real-time total input power of the pump group to obtain a product signal. The product signal is then subjected to low-pass filtering to remove high-frequency disturbance components and obtain the corresponding DC component. The positive and negative signs of this DC component are used as the direction component and the magnitude is used as the amplitude component, which together constitute the pump group power variation component corresponding to the distribution pipeline. A reverse opening correction amount is generated based on the direction component, and the correction amount amplitude is determined according to a preset opening amplitude threshold. At the same time, the demand value corresponding to the distribution pipeline is obtained, and the real-time flow rate of the distribution pipeline is compared with the demand value. If the real-time flow rate is lower than the demand value, the correction direction is forced to the direction of increasing the opening. If the real-time flow rate meets the demand, the gradient reverse correction direction is retained, and finally, the opening correction amount after flow constraint adjustment is output. In a preferred embodiment of the present invention, the process of performing phase-sensitive demodulation on the sinusoidal reference signal and the real-time total input power includes: The real-time total input power is multiplied by the sinusoidal reference signal to obtain a product signal; the product signal is low-pass filtered to remove components with frequencies exceeding the preset micro-amplitude value to obtain a DC component; the sign of the DC component is obtained and recorded as the direction component; the amplitude of the DC component is obtained and recorded as the amplitude component; the pump power variation component is composed of the direction component and the amplitude component. Understandably, the total power signal of the pump unit contains small fluctuations caused by the micro-vibrations of all valves, similar to multiple weak AC signals of different frequencies superimposed on a large DC bias. To extract the impact of the opening fluctuation of a certain electric valve on the total power, the unique frequency of that electric valve is used as a reference signal and multiplied with the total power signal. According to the orthogonality of trigonometric functions, only the component with the same frequency as the reference signal will produce a DC term after multiplication, while the product of all other components with different frequencies will fall in the AC frequency band. After low-pass filtering removes all AC components, the sign of the remaining DC component directly reflects whether the increase in the opening of this electric valve leads to an increase or decrease in the pump unit power. In a preferred embodiment of the present invention, the process of determining the amplitude of the opening correction includes: Obtain the amplitude components of each transmission and distribution pipeline, select the amplitude component with the largest value, and record it as the maximum amplitude component; for any transmission and distribution pipeline, obtain the ratio of the amplitude component of the transmission and distribution pipeline to the maximum amplitude component, multiply the ratio by the preset micro-vibration amplitude value to obtain the correction amplitude, and set the correction amplitude as the opening amplitude threshold. It is understandable that the amplitude component reflects how drastic the change in the opening of the electric valve is on the power of the pump group, i.e. the magnitude of the gradient; the farther away from the optimum, the steeper the gradient and the larger the amplitude; when approaching the optimum, the gradient is gentle and the amplitude tends to zero. The amplitude components of each branch are normalized to the maximum amplitude components in the transmission and distribution pipelines of all branches, so that the step size is dynamically adjusted between 0 and the preset micro-amplitude value. The purpose is to take large steps and fast speed in the early stage of iteration to accelerate convergence, and to make small steps and fine adjustments when approaching convergence to avoid overshoot and oscillation. In a preferred embodiment of the present invention, the demand value of the distribution pipeline is specifically the real-time demand chilled water flow rate value output by the terminal device corresponding to the distribution pipeline after closed-loop control calculation based on the deviation between the current temperature and the set temperature of its service area; the demand value is reported by the terminal device through the building automation network according to a preset communication cycle, and the previous reported value remains unchanged when no new reported value is received; In a preferred embodiment of the present invention, the process of adjusting the opening correction amount based on the comparison results includes: If the real-time flow is less than the demand value, the direction of the opening correction amount is adjusted, and the opening correction amount after the direction adjustment is recorded as the adjusted opening correction amount; if the real-time flow is greater than or equal to the demand value, the opening correction amount is retained and directly recorded as the adjusted opening correction amount. The process of adjusting the direction of the opening correction amount includes: If the direction of the opening correction amount is the direction of decreasing opening, then the direction of the opening correction amount is adjusted to the direction of increasing opening; if the direction of the opening correction amount is the direction of increasing opening or is zero, then it remains unchanged. Understandably, if the direction is pointing towards closing, but the flow rate is already insufficient, continuing to close the flow rate will only exacerbate the flow deficit, so it must be reversed, i.e., changed to opening. If the direction is already pointing towards opening, it means that the system is in a favorable state where opening can alleviate the insufficient flow rate and reduce pump power, so there is no need to intervene and the original direction should be maintained. If the correction amount is zero, i.e. the gradient is too small or has converged, there is also no need to intervene. In a preferred embodiment of the present invention, the process of obtaining the adjusted opening correction amount further includes: The amplitude of the adjusted opening correction is compared with the preset maximum step size of a single cycle. If the amplitude of the adjusted opening correction is greater than the maximum step size of a single cycle, the amplitude of the adjusted opening correction is truncated to the maximum step size of a single cycle, while the direction remains unchanged. Understandably, while the dynamic step size strategy accelerates convergence, in certain extreme conditions, such as when the system is just starting up or when the hydraulic state changes drastically, the demodulated amplitude may be abnormally large, leading to an excessively large single opening correction step size, causing drastic fluctuations in the terminal flow or even the risk of water hammer. Setting a maximum single-cycle step size as a safety upper limit cuts off the correction amount that exceeds the limit, which is equivalent to introducing a gradient pruning mechanism in the gradient descent iteration. This retains the characteristics of fast convergence while preventing system instability caused by excessively large single steps. Step S3: For any distribution pipeline, determine in real time whether the adjusted opening correction amount exceeds the preset convergence threshold. If it exceeds the preset convergence threshold, adjust the opening of the electric valve of the distribution pipeline according to the adjusted opening correction amount. In a preferred embodiment of the present invention, the process of adjusting the opening degree of the electric valve in the distribution pipeline includes: The system obtains the current real-time total flow and real-time total head of the pump unit. When the real-time total head is lower than the preset minimum safe head, the power supply frequency of the pump unit is increased. When the real-time total flow exceeds a preset proportion of the sum of the demand values ​​of each transmission and distribution pipeline, the power supply frequency of the pump unit is decreased. It is understandable that when the electric valves of each distribution pipeline gradually release excess resistance and the total system impedance decreases, if the pump frequency remains unchanged, the pump's operating point will move to the lower right along the characteristic curve, which may result in excessively low head or excessively high flow. Therefore, a coordination mechanism is needed. When the total head is below the safety threshold, it indicates that the pump's operating point has deviated from the effective range, and the frequency must be increased to compensate for the head. When the total flow exceeds a certain proportion of the sum of all terminal demands, it indicates that the pump speed is too high, and the frequency can be reduced to directly save energy and reduce flow redundancy. In a preferred embodiment of the present invention, the convergence threshold is 10% and 30% of the preset micro-amplitude value; This step is executed cyclically within the operating cycle of the multi-pump standby low-resistance distribution system. The intelligent control unit verifies the opening correction value of each distribution pipeline after adjustment at preset control intervals and compares it with the preset convergence threshold. When the correction value exceeds the convergence threshold, the opening correction value is added to the current steady-state opening command of the corresponding electric valve to complete the single-cycle opening fine-tuning. While adjusting the electric valve opening, the real-time total flow and real-time total head parameters of the pump group are collected simultaneously. When the total head is lower than the preset minimum safe head, the operating frequency of the multi-pump standby pump group is increased. To ensure distribution pressure; when the total flow exceeds the preset ratio of the sum of the demand values ​​of each branch, the pump group operating frequency is reduced to reduce redundant distribution energy consumption; during multiple rounds of iterative adjustment, the electric valves of the distribution pipelines of each branch gradually approach the optimal opening position, and the pump group operating parameters are synchronously matched with the real-time resistance characteristics of the system, so that the entire system converges to the operating state that meets the flow demand of all branches and the total power of the pump group is at a local minimum. Without changing the low-resistance distribution hardware architecture, ineffective throttling losses are further eliminated, and the variable flow adaptability and overall operating energy efficiency of the multi-pump standby system are improved.

[0019] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the present invention should still fall within the scope of the present invention.

Claims

1. A pump set energy consumption optimization method based on machine learning, characterized in that, Includes the following steps: Step S1: Obtain all the transmission and distribution pipelines of the pump group, and obtain the electric valves of each transmission and distribution pipeline. For any electric valve, generate an opening micro-vibration signal for the electric valve. The amplitude of the opening micro-vibration signal is a preset micro-vibration amplitude value, and the frequency is different from that of the other electric valves. Step S2: Obtain the real-time total input power of the pump set. For any distribution pipeline, obtain the real-time flow rate of the distribution pipeline and generate a sinusoidal reference signal with the same frequency as the opening micro-vibration signal of the distribution pipeline. Perform phase-sensitive demodulation operation on the sinusoidal reference signal and the real-time total input power to obtain the pump set power variation component of the distribution pipeline. The pump set power variation component includes a direction component and an amplitude component. Based on the direction component, an opening correction amount is generated with a direction opposite to the direction component and an amplitude of a preset opening amplitude threshold. Obtain the demand value of the transmission and distribution pipeline, compare the real-time flow with the demand value, and adjust the opening correction amount according to the comparison result to obtain the adjusted opening correction amount; Step S3: For any distribution pipeline, determine in real time whether the adjusted opening correction amount exceeds the preset convergence threshold. If it exceeds the preset convergence threshold, adjust the opening of the electric valve of the distribution pipeline according to the adjusted opening correction amount.

2. The pump group energy consumption optimization method based on machine learning according to claim 1, characterized in that, In step S1, the generation process of the micro-vibration signal of the opening degree of each electric valve includes: A fundamental frequency is set, and the frequencies of the micro-vibration signals for the opening degree of each electric valve are configured to be integer multiples of the fundamental frequency, with the integer multiples for each electric valve being coprime; the amplitude when the electric valve is fully open is obtained and denoted as the maximum amplitude F. max The preset micro-amplitude value is configured to be within the range of [1%F]. max 5%F max ].

3. The pump group energy consumption optimization method based on machine learning according to claim 1, characterized in that, In step S2, the process of performing phase-sensitive demodulation on the sinusoidal reference signal and the real-time total input power includes: The real-time total input power is multiplied by the sinusoidal reference signal to obtain a product signal; the product signal is low-pass filtered to remove components with frequencies exceeding the preset micro-amplitude value to obtain a DC component; the sign of the DC component is obtained and recorded as the direction component; the amplitude of the DC component is obtained and recorded as the amplitude component; the pump power variation component is composed of the direction component and the amplitude component.

4. The pump group energy consumption optimization method based on machine learning according to claim 1, characterized in that, In step S2, the process of determining the magnitude of the opening correction includes: Obtain the amplitude components of each transmission and distribution pipeline, select the amplitude component with the largest value, and record it as the maximum amplitude component; for any transmission and distribution pipeline, obtain the ratio of the amplitude component of the transmission and distribution pipeline to the maximum amplitude component, multiply the ratio by the preset micro-vibration amplitude value to obtain the correction amplitude, and set the correction amplitude as the opening amplitude threshold.

5. The pump group energy consumption optimization method based on machine learning according to claim 1, characterized in that, In step S2, the process of adjusting the opening correction amount based on the comparison result includes: If the real-time flow is less than the demand value, the direction of the opening correction amount is adjusted, and the opening correction amount after the direction adjustment is recorded as the adjusted opening correction amount; if the real-time flow is greater than or equal to the demand value, the opening correction amount is retained and directly recorded as the adjusted opening correction amount.

6. The pump group energy consumption optimization method based on machine learning according to claim 5, characterized in that, In step S2, the process of adjusting the direction of the opening correction amount includes: If the direction of the opening correction amount is the direction of decreasing opening, then the direction of the opening correction amount is adjusted to the direction of increasing opening; if the direction of the opening correction amount is the direction of increasing opening or is zero, then it remains unchanged.

7. The pump group energy consumption optimization method based on machine learning according to claim 1, characterized in that, In step S2, the process of obtaining the adjusted opening correction amount also includes: The amplitude of the adjusted opening correction is compared with the preset maximum step size for a single cycle. If the amplitude of the adjusted opening correction is greater than the maximum step size for a single cycle, the amplitude of the adjusted opening correction is truncated to the maximum step size for a single cycle, while the direction remains unchanged.

8. The pump group energy consumption optimization method based on machine learning according to claim 1, characterized in that, In step S3, the process of adjusting the opening degree of the electric valve in the distribution pipeline includes: The system obtains the current real-time total flow rate and real-time total head of the pump set. When the real-time total head is lower than the preset minimum safe head, the power supply frequency of the pump set is increased. When the real-time total flow rate exceeds a preset proportion of the sum of the demand values ​​of each transmission and distribution pipeline, the power supply frequency of the pump set is decreased.

Images